These features were previously experimental (hidden behind a Labs toggle) and are now available by default. Each also has its own detailed page in Omniscio's help library.

• Clipboard history — a clipboard-history button in the chat composer that keeps your recent copies (text + images) so you can re-insert them.
• Inbuilt terminal — a real shell terminal inside Omniscio, scoped to a project (xterm.js + node-pty), with Terminal command blocks that mark each command's exit code and add fold, copy, and re-run per block.
• Ask about this page — a floating button (and Ctrl+J) opens a popover that asks Claude about whatever is on screen.
• Weekly Session Analysis — a developer-tools panel that mines your past agent sessions for friction patterns, with an opt-in weekly deep-analysis report.
• Dev Pipeline — the hubs-bar panel that oversees the dev-pipeline workflow — see every active run, control gate auto-approval, and watch the auto-lander land ready branches.
• Native slash commands — type /model, /clear, /cost, /context, and /status in the composer and have them run for real in Claude Code.
• Trim unused snooze times — drops the snooze durations you never pick so the menu shows only the ones you use.
• Share Comments — viewers can leave inline or general comments on your shared links, cross-posted to Team Chat and your inbox.
• Custom Providers (bring your own web AI provider) — add an OpenAI- or Anthropic-compatible endpoint and run it as a named provider in the Claude Code harness.
• Portable Backup — export one encrypted backup file you can carry to another computer, and import it back.
• Pre-send spend warning — warns you before you send a prompt estimated to cost more than a threshold you set.

A tour dims the rest of the app, cuts a spotlight around the element it is describing, and shows a short card beside it explaining what that control does. Welcome and completion steps are full-screen cards instead. Tours are opt-in walkthroughs — nothing ever starts one for you.

Taking a tour

1. Open Settings → Setup Wizards. Every available tour is listed, with a checkmark beside the ones you have already finished.
2. Click a tour to start it. Use Next and Back to step through; the dots at the top of the card show how far along you are.
3. Press Esc to leave at any point, or click Skip — either one marks the tour finished, so it shows a checkmark on your next visit.

A few tours are offered rather than listed: when you first switch on a feature such as Gmail, Voice Commands or Screen Capture, a toast offers its tour. Accept it, or dismiss it for good with Never show again.

Show me around the app

The first-run walkthrough is the Interactive Demo. It runs on a sandboxed sample workspace with made-up hubs and sessions, so you can watch how the dashboard, the inbox and a conversation behave without touching anything real. It plays once during first-time setup, and you can run it again any time from Settings → Setup Wizards. Some of its steps are hands-on: instead of pressing Next you perform the real action, and the tour moves on when you do. Because the demo is sandboxed, nothing you do there creates anything.

Managing Your Sessions, launched from the Tours menu in the top toolbar, works the same way on a sample workspace — walking you through pausing, snoozing, undoing and archiving while showing each result and its shortcut.

Settings → Setup Wizards gathers every guided setup and walkthrough in the app onto one page. (Settings search finds it under “Setup Assistant” too.) Nothing here runs by itself — the page only starts something when you click it.

What is on the page

• Setup Wizard — walks the first-launch setup again: account, toolchain, theme and mobile access.
• Interactive Demo — the sandboxed walkthrough of the dashboard, with sample hubs and sessions. See App Tour.
• Onboarding Tour — replays the guided welcome tour: the inbox, replying to a session, the daily briefing, and finding your spend. Shows a green check once you have played it.
• App Tours — the full list of guided tours, each with a checkmark when completed.
• Vault Quick Reference — take or retake the knowledge vault's first-time card. Marked with a checkmark if you have done it before.
• Writer Studio Setup — the first-time setup for Writer Studio, replayable on its own.

System Requirements

Installing Omniscio

Omniscio ships in three Windows formats — a standard installer plus two portable builds. You'll get the ones your distribution channel offers:

1. Installer .exe (Omniscio-Setup-<version>.exe) — the standard Windows setup. It creates Start-menu and desktop shortcuts and updates itself in place, so it's the easiest way to stay on the current version. Like the portable builds it is code-signed as JLS Publishing LLC.
2. Single-file .exe (Omniscio-Portable-<version>.exe) — double-click and it runs; nothing to unzip. The trade-off is a slower startup: it unpacks itself to a temp folder on every launch (see Startup Splash below). Best when you just want one file to hand to someone.
3. Unzip-and-run .zip (Omniscio-<version>-unzip-and-run.zip) — extract it once to a folder, then run Omniscio.exe inside. It launches instantly with no per-launch unpacking. Best for daily use.

Whichever you pick, Windows may still show a SmartScreen warning — the app is signed as JLS Publishing LLC, but a brand-new signing certificate takes time to build SmartScreen reputation. Click "More info" then "Run anyway." On first launch you'll be prompted to sign in — see Authentication for details.

Startup Splash (portable .exe)

Each time you double-click Omniscio Portable.exe, two things happen before the main window appears:

1. Unpacking (tens of seconds, every launch) — the single-file portable wrapper unpacks its ~1.6 GB payload into a temp folder. A static dark splash reading "Omniscio — Unpacking files, please wait — The app will open automatically" appears on your primary monitor. This happens on every launch, not just the first — it's the price of the single-file format. The unzip-and-run zip skips this step entirely and goes straight to Boot.
2. Boot (~2–10 s) — a small frameless splash window appears with the Omniscio logo and an indigo spinner, showing a 5-phase checklist — Loading settings → Preparing your data → Restoring your work → Signing in → Opening the app — where each phase shows a check when finished, a pulsing dot while active, and stays dim until reached. A live line underneath names the current step ("Applying the latest updates...", "Restoring previous sessions...") — during database updates it even counts them off ("Applying update 3 of 12...") — and nothing but real progress ever changes that line, so it always names the work actually being done. If a single step goes quiet for a while, a second, quieter line appears beneath it saying only that Omniscio is still working ("Still working — larger projects can take several minutes...") so a longer wait never looks frozen. It closes automatically the moment the main Omniscio window is ready.

If the boot splash ever gets in your way, set the env var AMC_DISABLE_SPLASH=1 before launch — Omniscio will start straight to the main window with no splash. A built-in 90-second watchdog force-closes the splash if startup hangs, so you'll never be stuck staring at a frozen "Loading..." box.

Prerequisites: Claude Code CLI

Omniscio uses Claude Code (Anthropic's command-line coding tool) under the hood to run AI sessions. You need to have it installed:

1. Open a terminal and run: npm install -g @anthropic-ai/claude-code
2. Verify it's installed: claude --version
3. Omniscio will automatically detect Claude Code and use it for sessions.

Omniscio isn't in the App Store or Google Play. Instead it installs straight from your own computer as a web app: you open it on your phone once, then add it to your home screen. The whole thing takes about a minute. There are two parts — (1) get Omniscio open on your phone, then (2) save it as an app.

Step 1 — Open Omniscio on your phone

1. On your computer, click the phone icon 📱 in Omniscio's toolbar (top-right corner).
2. Click Enable & Show QR. Omniscio shows a QR code (and a link you can copy).
3. Open your phone's camera and point it at the QR code, then tap the link that pops up.
4. Omniscio opens in your phone's browser, already signed in. Have a look around — this is the full mobile interface.

Step 2 — Save it as an app

With Omniscio open in your phone's browser, save it as an app. The steps differ slightly on Android and iPhone.

🤖 Android (Chrome)

1. With Omniscio open in Chrome, tap the ⋮ menu (three dots, top-right).
2. Tap Install app (on older versions of Chrome it's called Add to Home screen).
3. Confirm. Omniscio is now installed on your phone. Its icon shows up in your app drawer (swipe up from the bottom of the screen to see all your apps) — and on some phones it's added to your home screen automatically too.
4. Open Omniscio from the app drawer — it runs full-screen, with no browser address bar, just like a regular app. To keep it handy, press and hold its icon and drag it onto your home screen.

🍎 iPhone & iPad (Safari)

1. Open Omniscio in Safari — on iPhone/iPad this must be Safari, not Chrome.
2. Tap the Share button (the square with an up-arrow, at the bottom or top of the screen).
3. Scroll down and tap Add to Home Screen.
4. Tap Add (top-right). An Omniscio icon appears on your home screen and opens full-screen.

Use it away from home (optional)

By default the phone app works on your home Wi-Fi. To use it from anywhere — on cellular, at work, travelling — turn on Access from anywhere in the same phone dialog. Omniscio offers to install Tailscale (a free, secure-connection app) and walks you through a one-time sign-in. After that, the installed app keeps working off your home network too. Full details and troubleshooting are in Mobile Access.

Omniscio supports two ways to authenticate with Anthropic's AI service:

Option 1: OAuth Login (Recommended)

This is the easiest way to get started. It uses the same login you'd use on console.anthropic.com.

1. Click "Sign In" on the authentication screen.
2. A browser window opens to Anthropic's login page.
3. Sign in with your email, Google, or GitHub account.
4. Authorize Omniscio to access your account.
5. You're redirected back to Omniscio — you're all set!

Option 2: API Key

If you prefer, you can use an Anthropic API key directly.

1. Go to console.anthropic.com/settings/keys and create a new API key.
2. In Omniscio, choose "Use API Key" on the auth screen.
3. Paste your API key and click "Save."

Multiple Accounts

You can add multiple Anthropic accounts to Omniscio. This is useful if you have a personal account and a work account. Omniscio will automatically switch between them for rate-limit recovery — if one account hits its usage limit, Omniscio can continue working with another. When every usable account is at its limit, the affected sessions show the orange Rate Limited (Waiting) state — not the amber “Needs You” state — write a “Waiting for account capacity” note in their chat, and resume automatically when a limit resets or an account frees up. No action is needed from you.

Distribute Sessions Across Accounts goes a step further: instead of piling every session onto one account until it caps, Omniscio spreads new sessions across all your login accounts by usage — with a gentle preference for an account whose usage window is about to reset — weighted most toward the weekly (7-day) window, since a whole week's unused budget is the bigger thing to lose — so capacity that would otherwise expire unused gets used first. It also caps how many sessions stack on any one account, so new work fans out to accounts with room instead of a single account running away with the fleet. With three accounts you get roughly three times the effective 5-hour window, each account drains slower, and there's no synchronized "switch everything at once" stall — when one account caps, only that account's sessions move. Each session otherwise stays on its own account. It is off by default — turn it on at Settings → Accounts (the setting appears once you have 2+ login accounts). Until you do, every new session lands on the single active account; switching it on turns on the fan-out described here, and switching it back off restores the single-active-account behavior exactly. When it's on, the account pill at the top of the window shows a balance-scale icon with the number of accounts currently working, plus a green/amber/red health mark for whether your whole pool is on track to last until reset — and opening it tags each account with how many of its sessions are actively running versus parked waiting on you (for example "7 running · 6 waiting") — so a big count next to low usage isn't a mystery, since the parked ones aren't burning anything. An account that's maxed out instead shows a single muted "waiting" tag, since those sessions can't run until it frees up. If one account gets busy — or is simply holding far more sessions than its fair share — while others sit idle, Omniscio also nudges a couple of its sessions onto a freer account before they stall (when the account isn't near its limit it moves only idle sessions, so nothing you're actively working on is interrupted), and parks a session that keeps bouncing across a fully-maxed pool until capacity returns. On a phone, the same balancing indicator appears on the account button in the mobile header.

3

Credential Security

Your credentials are stored securely on your computer using your operating system's own keyring — Windows DPAPI, macOS Keychain, or Linux libsecret. Omniscio never transmits them to any server other than Anthropic's API — nothing sends them anywhere on its own.

The one way credentials can travel is if you deliberately carry them, and both routes are opt-in and encrypted. Ticking Include my saved credentials on a Portable Backup re-encrypts your API keys and sign-in tokens into that one file so they can be restored under the new computer's own secure storage — it is off by default, so an export without it carries no secrets at all. And Backup Mirror copies your config to a folder you choose (Dropbox, OneDrive, iCloud); encrypted credentials stay encrypted in the mirror.

Because DPAPI keys are tied to your Windows install and hardware, a major hardware change (swapping your CPU or motherboard) can reset them, after which Omniscio can no longer decrypt accounts you signed in before the change. When that happens Omniscio sets the affected account aside instead of looping on it, and shows a "Login Expired — sign in again" notice for that specific account — just re-sign in at Settings → Accounts and it picks back up right away, no restart needed. The same applies to a saved Anthropic API key: its card under Settings → Accounts turns amber "Can't be read — re-enter" instead of a false green — the key is still valid, Omniscio just can't read its encrypted copy — so paste the same key back into that card and it's repaired in place. You don't have to go hunting for it, either — when a saved API key can't be read, Omniscio also drops a notice in your Inbox the next time you open the app, pointing you to re-enter it; re-entering clears it.

Database Encryption

Beyond your credentials, Omniscio can encrypt your entire local database at rest — mission-control.db and its cold-storage archive — so the files on disk are unreadable ciphertext without your key. It's off by default (if you never turn it on, nothing changes), and you switch it on from its own Database Encryption settings page — it has no sidebar row, so see the Database Encryption guide for the way in. A random key is sealed in your computer's keychain (or, for zero-knowledge, a passphrase you set), and enabling it re-encrypts your existing database crash-safely on the next restart.

Buy Credit

You can top up your prepaid gateway AI credit with a credit card straight from Omniscio. The Buy Credit card lives on the one money screen — Settings → Plan & Usage — directly under your Credits remaining balance, so adding more is right where you check what's left. (Settings → My API Keys shows the same balance but has no purchase control; it just links you across to Plan & Usage.)

1. Open Settings → Plan & Usage and find the Buy Credit card under your Credits remaining balance.
2. Pick an amount — a one-tap preset ($10 / $25 / $50 / $100) or a custom amount from $5 to $1,000.
3. Omniscio opens Stripe's hosted Checkout page in your browser. You enter your card on Stripe's own page — card details never go through Omniscio.
4. Finish paying and return to Omniscio. Your balance updates shortly after — if you don't see it right away, refresh the balance.

Buying credit is 1:1 face value: pay $25 and you get $25 of credit. The card processing fee is absorbed — it's not added on top and it doesn't shrink the credit you receive.

Plan & Usage

Omniscio keeps everything about your plan and spending on one screen — Settings → Plan & Usage (renamed from "Plan & Billing" and expanded in 2026 — the screen is the Plan & Billing (Pro subscription) screen). See the plan you're on and this month's AI allowance (it resets at the start of each month), see and top up your prepaid credit balance, compare Free / Pro / Team and upgrade or manage your subscription (Pro is $15/mo through Stripe), and bring your own API key. Your recurring subscription and your prepaid Buy Credit balance are still separate money paths that never cross — you just see and manage both in one place here.

When you open the app, Omniscio shows a "Sign in to continue" screen. This is the Omniscio sign-in — it's separate from how you connect to Anthropic's AI service (covered under Authentication). Sign in with Google to continue into the app — signing in is required to use Omniscio, and it's what unlocks features like running recipes.

Signing in

• A verified email is required for new accounts — you confirm your email address before the account is active.
• Once you're signed in, Omniscio remembers it on your device, so you won't have to sign in again every time you open the app (you stay signed in even when you're briefly offline).
• You can sign out at any time from Settings → Accounts; the sign-in screen reappears the next time it's needed.

Self-serve workspaces

When your install uses the optional cloud sign-in (Global Auth), an eligible signed-in user on a paid plan can create and own their own workspace — a separate organization with its own members — and belong to several at once. A workspace switcher in the User Management header moves you between the workspaces you belong to; switching re-loads that workspace's data without signing you out on your other devices. If you're on the free plan and try to create a workspace, Omniscio sends you to the Plan & Usage screen to upgrade. A workspace owner or admin also gets a scoped view of User Management for managing just their own workspace's members. This is an owner/admin feature and appears only when cloud sign-in is turned on.

Account deletion & your data

You can delete your Omniscio sign-in account yourself, from inside the app — you don't need to email anyone or wait for someone to action it.

1. Open Settings → Accounts.
2. Choose Delete my account and confirm.
3. Your account is removed immediately — this is a permanent (hard) delete of the sign-in account, including the cloud sign-in record. It is not a "soft delete" you can undo.
4. The data on this computer is cleared the next time you open Omniscio, and you land on a fresh setup. It can't happen instantly because the local database is open and in use while the app is running.

What the app operator can see about your account

When sign-in is active, the person running this Omniscio install can see, for a signed-in user, your account email, display name, and plan/tier — for example, attached to a support, help, or bug-report message so they know who it came from. Your sign-in tokens (the secret part of logging in) are never shared this way. When no one is signed in, those messages are anonymous.

Asking us to delete your data

Besides deleting your own account, you can request deletion of data held about you. How requests are handled — and the rights you have — will be set out in the Privacy & Your Data policy when it's published.

Omniscio keeps a working copy of much of your data — your projects, sessions, chat history, and messages from connected channels — in a database on your own computer. As a multi-user cloud product it also keeps your account and some synced data on our servers, and several features send data to outside services to work (for example, an AI model needs your text to summarize it). In plain language, here’s what that means:

• What personal data is stored, and where — in the local database on your machine, and on our servers for your account and synced data.
• Which outside services receive data when a feature runs — AI models, voice, email, crash reporting, and so on — and what each one gets.
• How long data is kept, and how to remove it.
• Your choices and rights — including turning features on or off, deleting your account (see Account & Sign-In), and requesting deletion of data held about you.

Health & error reporting

To catch crashes and bad releases across everyone's installs, Omniscio sends a small, anonymous stream of usage and error health data. It is always on — there is no setting that turns it off; the disclosure lives at Settings → System on the “Usage telemetry (always on)” card, which says so in as many words. What you can control is Auto-Email Crash Reports at Settings → System → Auto-Email Crash Reports: it decides whether a crash also emails a scrubbed log excerpt, and turning it off does not stop crash capture itself (a restart applies the change).

What's sent — all tied to a random per-install ID, never your name or machine:

• Crash and error signals — the first line of the message, with secrets and personal details scrubbed out and truncated.
• A once-a-day health summary — which features were used and how often, uptime, error/warning counts, session stats, and your total AI spend. All counts, never content.
• Anonymous automation reliability counts (recipe supervisor, PM automations, and CLI failures — as fixed codes and counts, never the content) and feature-adoption counts (voice and saved-browser-login usage — counts only, never transcripts or login names).
• App version and platform (e.g. win32).

If you're signed in, the same daily numbers (your AI spend and usage counts) are also tagged with your user ID, so the maintainer's admin console can show per-user totals. Still counts only — never your name, email, messages, or files. Deleting your account removes them.

Never sent: your username, file paths, project names, message or conversation content, API keys, or account emails.

Data erasure (F033)

When you delete your account, Omniscio performs a comprehensive, automated data erasure across every surface where your data is stored:

• Team Chat messages — all message bodies you authored are redacted and replaced with a deletion tombstone, while the conversation structure stays intact for other members.
• Search index — every searchable copy of your messages is deleted so your content no longer appears in search results.
• Device tokens and preferences — your push notification tokens, notification preferences, presence data, and saved items are hard-deleted.
• Direct messages — your PII (display name, email, photo) is scrubbed from connection records.
• Denormalized fields — channel previews showing your name are cleaned.
• Scheduled messages — any pending or dead-lettered scheduled messages you authored are deleted.

A daily reconciler automatically re-drives any residual erasure that could not complete in a single pass (for example, for very prolific users), so nothing is left behind.

Omniscio's enterprise identity features are per-organization: your settings, your SCIM token, your claimed domain, independent of every other organization.

Domain verification

Organizations can verify and claim their email domain. Verification uses a DNS TXT record challenge — you publish a verification token in your DNS, and Omniscio confirms ownership before claiming the domain for your organization.

• Domain verification — prove ownership with a DNS TXT record.
• One org per domain — domains are transactionally claimed to prevent conflicts.
• Per-org configuration — each organization's identity settings are independent of every other organization's.

SCIM v2 provisioning

SCIM v2 (System for Cross-domain Identity Management) lets your identity provider automatically create, update, and deactivate user accounts in Omniscio. Each organization gets its own SCIM endpoint.

• Automated provisioning — new hires get access automatically when added in your IdP.
• Automated deprovisioning — departing employees are removed when disabled in your IdP.
• Per-org SCIM token — your organization gets its own bearer token, scoped to your org alone (stored hashed, plaintext shown to you once when it is minted). Omniscio mints and rotates it for you on request — see below.
• Per-org endpoint — /scim/v2/{orgId}, speaking SCIM 2.0 over the Users resource (see the supported subset below).

The dashboard is the main screen you see when you open Omniscio. It's divided into three main areas:

4

Left Sidebar: Hubs & Sessions

The left side shows all your hubs (folders on your computer that you've added to Omniscio). Under each hub, you can see its sessions (active AI conversations). Click a session to open its chat in the center panel.

At the top of the sidebar, you'll see tabs to switch between:

• Regular view — Your hubs organized by folder, with sessions underneath
• Inbox view — Only sessions and items that need your attention

Center: Session Chat

The center area shows the conversation with the currently selected session. This is where you read what the AI agent is doing and type your responses. See Chat Interface for details.

Top Toolbar

The toolbar at the top contains quick-access buttons for various features:

Built-in sidebar integrations (such as Gmail, Stats, or Mind Map) and installed third-party plugins that have a sidebar panel also get a toolbar button automatically. These appear in the … (overflow) menu by default; right-click any item to pin it to the visible toolbar.

The header's right side is a row of draggable blocks: the whole shortcut bar is one block, and the account / balancer and Hardcore widgets sit as their own blocks beside it. Drag a widget block directly (a quick tap still opens its panel), or drag the shortcut bar by the small grip that appears when you hover it, to rearrange them — your layout is saved and survives restarts. Shortcut icons still reorder inside the bar, and Settings → Widgets controls which widgets show or hide.

System Tray

Omniscio also lives in your Windows system tray (the small icon area near the clock). The tray icon shows a badge count of sessions that need your attention. Left-click the icon to show / focus the window. Right-click for the menu:

• Show Mission Control — brings the window to the front (same as a left-click on the icon)
• Pause Wake Word / Resume Wake Word — only appears when wake-word voice activation is enabled in Settings → Voice Control
• Hard Reload — clears the renderer cache and reloads the window. Use it if the UI looks stuck or didn't pick up a change. Claude sessions keep running in the background — only the UI reloads.
• Restart Mission Control… — asks for confirmation, then relaunches Omniscio. Running Claude sessions are stopped gracefully (same as Quit) and will not auto-resume on the next launch — each one gets a “Session suspended — app was closed” marker. Use it after changing a setting that needs a restart, or if the app feels wedged.
• Open Developer Tools — opens Chrome DevTools for the Omniscio window in a separate detached window. Mostly useful for debugging or reporting issues.
• Quit — gracefully shuts down Omniscio: terminates every Claude CLI process, force-kills anything still running after 5 seconds, then exits

Detached Session Window (Pop Out)

If you want a single session always visible on a second monitor while the main Omniscio window stays on its primary view, you can pop the session out into its own floating window. Open the session's ··· (overflow) menu in the session toolbar and pick Pop Out to Window — no setting to toggle, it's available on every session.

• The floating window shows the same chat, composer, and toolbar as the inline panel — there is only one conversation, both windows are views onto it.
• In the main window, the session's sidebar row shows a small "open in new window" chip, and the panel area is replaced by a placeholder with Bring window to front and Reattach here buttons. The session's ··· menu also swaps to Focus Detached Window while it's popped out.
• Omniscio remembers where you put each session's window and how big you made it — re-popping the same session restores that position and size.
• Closing the floating window (or clicking Reattach here) re-mounts the session in the main window.
• Pressing Ctrl+W / Cmd+W in the popped-out window opens a quick confirm — Close window (leave the session running in the background), Close & stop session (stop it too), or Cancel. The title-bar X still closes right away.
• If a session ends or is archived while popped out, the window stays open so you can keep reading the transcript — close it when you're done.

Every other pop-out window closes on Ctrl+W. A hub or app popped out into its own window (SuperMail, Tasks, Calendar, …), The Vault, Scratchpad, Writer, Support Chat, the Job Monitor and project docs all close on Ctrl+W / Cmd+W, exactly like their close button. If a dialog is open inside the window, the first Ctrl+W closes the dialog and the next one closes the window. It works while you are typing too, and a SuperMail reply you were writing is kept as a draft.

Reporting a Bug or Sending Feedback

The 💬 Feedback button in the top toolbar opens a short form: pick whether it's a bug, a feature idea, a question or general feedback, describe it, and optionally leave an email address so someone can reply to you.

• You get a reference number. After the report sends, the confirmation shows a short code such as ref: a1b2c3d4. Keep it. If you ever need to follow up, quoting that code is the fastest way for us to find your exact report.
• You can attach up to 5 files, 6 MB in total. Screenshots are the most useful thing to attach. If a file is too big, or the files together go over the limit, Omniscio tells you straight away rather than failing later.
• Some diagnostics ride along, and you can see them first. The form shows exactly what will be included before you send: app and platform versions, how long Omniscio has been running, how many sessions are active, and counts of any recent errors or crashes. Contact details and other sensitive text in your message are stripped out before it leaves your computer.
• Nothing is lost if you're offline. A report that can't be sent is saved on your machine and re-sent automatically later. If that happens you'll see a message saying so, with the same reference number.

Replies to your reports

Every report you send also opens a conversation with the support team, so they can answer you right inside Omniscio — you don’t need to turn on Get Help for this.

• You’re told when someone answers. A notification says An agent replied to your report, and the Feedback button shows a small dot until you read it. Click the notification to go straight to the reply.
• Find them under “Your reports”. Open the Feedback button and choose Your reports at the top of the form. The list shows each report you sent and where it stands: on its way, waiting on an agent, answered, or resolved.
• You can write back. While a report is open with the support team, type a reply at the bottom of it. Once it’s resolved it becomes read-only — send a new report if you need more help.
• A bug report or a feature idea goes to a person. Send one and you get a single email thanking you and saying someone is looking into it — then a person on the support team answers you here. Nothing is started automatically to work on it in the background, and a question or a general note works the way it always has.
• Only people answer these. The AI assistant never replies to a report conversation, and the conversation carries only what your report already sent: its text, plus the name and email on your account.
• You can switch it off. Turn off Report conversations in Settings → Lab and new reports no longer open a conversation, and Your reports is hidden.

Hubs in Omniscio represent folders on your computer where you do your coding work. Each hub can have multiple AI sessions running at the same time.

Adding a Hub

1. Click the "+" button at the top of the sidebar (or use the context menu).
2. Choose "Add Hub" and browse to the folder on your computer.
3. The hub appears in the sidebar with its folder name.

Hub Settings

Right-click a hub in the sidebar to access options. The most-used actions sit at the top level; the rest are grouped into hover/click submenus (look for the › chevron) so the menu stays short:

• New Session — Start a new AI session in this hub
• Run Recipe — Submenu: run a multi-step recipe against this hub
• Edit Hub — Change the hub name, color, group, or isolation
• Edit Notes — Jot freeform notes about the hub
• Files — Submenu: Open Hub Docs, Show in Explorer, Copy Path
• Appearance — Submenu: detect / choose / remove the hub icon, and show or hide the git branch in session headers
• Pin to Top / Unpin — Keep the hub pinned above the rest of the list
• Remove — Remove the hub from Omniscio (doesn't delete files)

The Files and Appearance submenus only appear for real folder-backed hubs — virtual hubs (Gmail, Calendar, Settings, etc.) show just the flat top-level items.

Moving a Hub to a New Folder

If you move a hub's folder on your computer — or just want to point Omniscio at a different location — you don't have to lose your history. Use "Move to a new folder…" in the Edit Hub dialog (right-click → Edit Hub), or "Locate moved folder…" on the red "hub folder is missing" banner.

1. Open Edit Hub and click "Move to a new folder…" (just under the folder path).
2. Pick the new folder.
3. Omniscio re-points the hub to it — all your sessions and history come along automatically, because they belong to the hub, not the path. Sessions that are still running keep working where they started; only new sessions use the new folder. Nothing on disk is copied or deleted.

Merging two hubs. If a hub already exists at the folder you pick, Omniscio offers to merge the two instead: it shows how many sessions will move, takes a one-time safety backup of your data first, then combines everything into the hub already there and retires the duplicate. This is the clean way to fix a split caused by adding the same folder twice. The merge is all-or-nothing, so it can never leave your data half-moved.

Built-in hubs (Gmail, Calendar, Settings, etc.) can't be moved — they aren't real folders.

Hub Colors

Each hub gets a colored bar on the left side of its sessions. You can customize this color to visually distinguish your hubs at a glance.

Virtual Hubs

Omniscio includes several virtual hubs that aren't tied to a folder on your computer. These represent special features:

Repo Foundations

The Repo Foundations feature puts a Set up this repo card on the hub dashboard for every folder-backed hub (on by default). Clicking it runs the Repo Foundations skill, which installs a tailored documentation system and a Claude-rules baseline after showing you a preview to approve. The setup spawns a real paid session; you can turn the skill off any time in Settings → Features.

Groups

You can organize your hubs into groups — named, collapsible sections in the sidebar. To create one, click the + button at the top of the hubs sidebar and choose New Group, then give it a name. The same menu's other entry, Add Hub, creates a hub.

A group header shows how many hubs sit inside it. Click the header to collapse or expand the group, and use its controls to rename it, delete it, pin it to the top of the sidebar, or add a hub straight into it. Groups and hubs can both be dragged into the order you want.

Deleting a group removes the group, not the hubs inside it, and the toast it shows offers an Undo that puts the group back along with its hubs.

5

Clean Room is a special project in your sidebar (it sits alongside the built-in ones like Session Search — it isn't a folder on your computer). Every session you start inside it talks to a “vanilla” AI: a completely fresh version with none of your personal setup and none of the extra help Omniscio normally adds. It's like meeting the AI for the very first time, on a blank page.

Why would you want that? Because sometimes the AI does something surprising and you want to know why. Is it the AI itself — or is it one of your rules, your notes, a connected tool, or something Omniscio quietly adds behind the scenes? Clean Room is your control group: run the same request here with nothing added, and compare it to a normal session. It's also handy for testing a fresh prompt with zero head start.

Opening Clean Room

1. Look in the sidebar for the Clean Room entry (🧼) among your projects.
2. Click it to open the Clean Room start screen — a short explainer of what Clean Room is (what's turned off versus what stays), with an AI picker and a Start a Clean Room session button.

Starting a Fresh Session

1. On the start screen, leave the AI on Claude (or pick another — see below), then click Start a Clean Room session.
2. A blank session opens — just type your message and press Enter. The AI that answers is the fresh, blank-slate version, with no customizations and no added context.

You can run several Clean Room sessions at once, just like any other project.

Picking the AI

On the start screen, an AI picker lets you choose which AI answers before you begin. It's set to Claude by default, and you can leave it there. In this first version you can also pick one of the Claude-compatible models — DeepSeek, Kimi, GLM, or MiniMax. Whichever you pick, it's still the same fresh, stripped-back session — only the underlying AI changes. You can choose a different AI for each session.

What's Turned Off (and What Stays)

Here's exactly what a Clean Room session leaves out, and what it keeps.

Each session also starts in its own private room — a blank, per-session workspace, so one Clean Room session can never see another's files, and the whole area lives outside the app's data folder so nothing sensitive sits next to it. A Clean Room session is fenced to its room: if it tries to read a file or run a command outside the room, it's blocked outright (see “What to Expect” below). To give a session files, use Open room folder and drop them in — it can freely read anything inside its own room.

What to Expect

A couple of things are worth knowing so nothing surprises you:

• It's still Claude Code, not a “bare” AI. Clean Room removes the extra layers around the AI, but the AI keeps its own built-in identity and its own tools. That's on purpose — it's what makes it a fair comparison to your normal sessions.
• It's a control group with a filesystem fence, not a locked-down sandbox. The AI's own built-in tools are still live and work freely inside its own room. But if a Clean Room session tries to read a file or run a command outside the room — like your real settings elsewhere on the computer — it's blocked outright (not just a prompt you might approve by reflex), so it can't be walked out of its folder. The room area also lives outside the app's data folder, so even a clever escape lands next to nothing sensitive. “Clean” means a blank context (none of your customizations, none of Omniscio's layers); the block adds a real limit on what it can reach on its own.

A session is a single conversation between you and a Claude Code AI agent. Each session runs in the context of a specific project folder on your computer.

Creating a Session

1. Select a project in the sidebar.
2. Click the "+" button next to the project name, or right-click the project and choose "New Session."
3. Type your initial message (what you want the AI to do).
4. Hit Enter to start the session.

Session handoff

Session handoff rolls a long session over into a fresh one that's already caught up — a "Hand off to a fresh session…" item in the session ⋯ menu, plus a one-per-session note once a chat has used a lot of context (and the same note when it gives up). The note never spends; only confirming the handoff pays for the one summary call.

Browser Logins

Browser Logins let you log into a website once and have every agent you run reuse that login automatically — many at the same time — until the site signs you out. Open Browser Logins in the left sidebar (under Agent Tools), click Add a login, and a real Chrome window opens; sign in to any site the normal way (passwords, 2FA codes and email magic links all work) and close the window. Passkeys are the one exception — Chrome turns them off in any browser window that is being driven automatically, which is how this feature works, so use a password in that window instead. That login is saved on your computer only — no passwords are ever stored, just the logged-in browser session — and each agent drives its own private copy, so they never collide. During first-run setup Omniscio installs the small browser-automation tool this uses in the background, so it's ready when you are. Remove a saved login anytime and it's erased from your computer along with its saved session.

Session Statuses

Each session has a colored status dot that tells you what's happening:

In a project's sessions sidebar these are grouped top-to-bottom into Needs You (sessions waiting on you — a question or an approval to continue), Interrupted (sessions whose process stopped or that hit an error or stalled — shown with an ember warning icon; revive one by sending it a message), Live (actively working or waiting on you), and Paused. Each group has a header you can click to collapse it, and the Interrupted group only appears when “show ended sessions” is turned on.

Errored and stalled sessions go quiet (updated 2026-08-11): a session that hits an error or stalls no longer pops into your Inbox, adds a project badge, or plays a chime. Instead it settles into the Interrupted group above — just like a session that ended — so you can get to it when you're ready without being interrupted. (One exception: if the app is already quietly reconnecting a dropped session, it stays in Live with a “Reconnecting…” dot until it comes back.) If a session truly can't recover on its own, it still surfaces in your Inbox as “Recovery failed” so you never lose track of it.

6

Project Count Badges

These session statuses roll up into small colored counts shown next to each project name in the sidebar, so you can see at a glance what's happening across a whole project without expanding it. They appear most-urgent first:

Idle and finished sessions aren't counted, and a project with nothing going on shows no badge at all. The red count is split out from the amber one so a session that's waiting on you in a failed state never hides inside the “needs you” number. (Updated 2026-08-11: sessions that simply errored or stalled are no longer counted here, and no longer land in your Inbox or the tray badge — they settle quietly into the Interrupted group, like ended sessions, until you get to them.)

The same colored counts also ride on the “Sessions” tab of any feature panel that hosts AI agents — Tasks, The Vault, Mind Map, Flowchart, Whiteboard, and SMS — so you can tell an agent is running or waiting on you without switching to that tab. They follow the same rules: most-urgent first, only non-zero counts, and hidden while Focus Mode is on.

Managing Sessions

Right-click a session to see available actions:

• Pause / Resume — Temporarily stop the AI, then continue later
• Snooze — Hide the session from your inbox for a set time (15 min, 1 hour, etc.)
• Save — Bookmark a session to the collapsible Saved section in the sidebar. If a saved session later needs you, it pops back to Needs You until you handle it, then returns to Saved. Keyboard shortcut: B.
• Archive — Move the session out of your active view. You can still find it later.
• Move to — Re-file the session under a different project. Hover the row for a moment (or click it) and its submenu pops out — a searchable list of eligible projects appears; click one and the session moves immediately, with an Undo toast. Also available from the overflow menu at the top of an open session.
• End — Permanently stop the session
• Exports — A side menu for saving or sharing the transcript, grouped into sections by destination with dividers: Copy as Markdown to the clipboard and Copy since last compaction (just the most recent compaction onward — falls back to the whole thread if the session hasn't compacted yet); Export to Markdown (a .md file), Export to Word (.docx) and Export to PDF; Export raw thread, a troubleshooting file with every message the session stored, word for word, each labelled with whether you saw it, could have expanded it, or never could (it is unredacted, so treat it as sensitive); Export to Google Docs; and Share the conversation as a read-only link (see Session Sharing). Export to Word and Export to Google Docs are off by default — turn Word on in Settings → Sessions and Google Docs on in Settings → Google Workspace; until then they stay hidden so the menu stays short. Apart from Export raw thread, the exports are visible-only — they contain just your messages and Claude's prose (including intermediate narration like “Let me check this file…”); the mechanical ▸/← tool-call and tool-result lines are stripped, system rows are dropped, and turns that were only tool activity collapse to nothing rather than emit an empty ### Claude header. A single line at the foot of the menu shows the size of what you'd export — words, characters, and tokens (e.g. 1,234 words · 32k chars · ~7.1k tokens); “Copy since last compaction” shows its own smaller token count in parentheses when it differs, with the word/character breakdown on hover.
• Fork / Aside — Start a side conversation branching from this session (see Asides)

Snoozing Sessions

When a session needs your attention but you're not ready to deal with it yet, you can snooze it. The session disappears from your inbox and reappears after the snooze period ends. Available snooze durations:

• 15 minutes
• 1 hour
• 4 hours
• Tomorrow morning
• Custom time

For custom times, you can type natural language like in 2 hours, friday 9am, compact shorthand like 805p or 8p, or tomorrow 6:00. When you skip the AM/PM marker on a clock time, the parser never picks sleep hours (12 AM–6 AM): hours 1–6 always resolve to PM (so tomorrow 1:00 means 1 PM tomorrow); for hours 7–12 after a day or date the time is genuinely ambiguous, so the palette offers both readings as two pickable rows (thursday 8 → 8 AM and 8 PM, so you pick the one you meant); and the word tonight forces evening (tonight 9 → 9 PM). Type the meridiem explicitly (6am) if you really want an early-morning time.

Typing a plain number is expanded into every interpretation it could mean, so you pick instead of guessing. A 3–4 digit number like 930 is read as a clock time and a date: it lists 9:30 PM, 9:30 AM (an ambiguous hour offers both, soonest first), then Sep 30 (the 9/30 date), then 930 minutes. 1231 lists 12:31 PM, Dec 31, and 1231 minutes. An impossible clock or date is skipped — 931 offers no “Sept 31” — and a smaller number keeps the older behavior (9 → 9:00 clock plus N minutes/hours/days).

The box also understands a much wider range of everyday phrases: the clock before the day (5pm tomorrow, eod friday), relative anchors (2 hours from now, a week from friday), business days (3 business days, skips weekends), ordinal days (the 15th, 25th of march), holidays (christmas, thanksgiving, christmas 9am), bare months (june, early august), vague waits (soon, shortly, in a while), end-of-period phrases (eom, eoy, cob), mealtimes (lunch, dinner, after work), and leading filler words (on friday, remind me in 2 hours, wake me at 7 all strip the filler automatically). Typing next monday when the nearest Monday is only a day or two away offers both This Monday and The following Monday; typing by friday offers both Friday 5 PM (End of day) and Friday 8 AM (Morning).

Archiving Sessions

Archiving a session hides it from your regular view but doesn't delete it. You can find archived sessions by searching or through the session history. Middle-click a session on desktop to quickly archive it.

Guided Missions (Windows)

On Windows, Omniscio offers guided missions — short, coached experiences that run inside the regular session panel as interactive cards. Four missions are available:

• Computer health check — scans for cleanable space, desktop clutter, and duplicate photos, then offers a one-click reversible tidy-up. This mission launches automatically the first time you create a project after setup.
• Start my first project — takes an idea you have in your head and sets up a folder with a plain-language plan and a starter file you can open right away. All undoable — one button removes everything the mission created, leaving the rest of your computer exactly as it was.
• Teach Omniscio about me — a short interview that asks a few friendly questions (what to call you, what you're working on, how you like answers), then saves a tidy “About me” block into your Custom Agent Instructions so every future session starts already knowing you. Fully reversible — one tap removes the saved profile and restores whatever you had before.
• Set up my daily briefing — picks a morning hour for your once-a-day digest of Omniscio activity, emails, calendar, and chats, then switches it on. Re-running the mission at any time just updates the hour. Fully reversible — one tap turns the digest back off and restores your previous settings.

To run a guided mission, click the Run a guided mission button that appears in any blank session (a session with no messages yet) or as a secondary action on the project empty-state panel. A small picker lets you choose which mission to start. The mission runs right there in the session, walking you through each step with buttons — no typing required until you want to.

How “Teach Omniscio about me” works

When you choose this mission from the picker, Omniscio starts a short, friendly conversation — four light questions you can answer (or skip) at your own pace:

1. Kickoff — “Want me to learn a little about you?” → Yes, get to know me or Maybe later.
2. Interview — A few plain-English questions: your name or preferred nickname, what you’re working on, how you like answers. Each one is skippable.
3. Confirm — “Here’s what I’ll remember about you:” followed by a short summary. Tap Save it to apply, or Tweak something to go back.
4. Done — Confirms the save and points you to Settings → Agent Instructions if you ever want to edit the profile by hand. An Undo button is always shown and restores the exact text you had before.

The profile is stored inside a clearly-marked section of your Custom Agent Instructions field — it never overwrites anything you had already written there. Re-running the mission later updates the same section in place.

How “Set up my daily briefing” works

When you choose this mission from the picker, Omniscio walks you through four quick steps:

1. Kickoff — “Want a friendly summary of your day each morning?” → Yes, set it up or Maybe later.
2. Pick a time — Choose one of the suggested slots (Early bird 6 am, Morning 8 am, Midday 12 pm) or enter any hour you prefer.
3. Confirm — “Your briefing will be ready around <time> in Briefings…” → Turn it on to apply, or Tweak to go back and change the hour.
4. Done — Confirms the digest is on and points you to the Generate now button if you want to see a sample straight away. An Undo button is always shown and restores the exact settings you had before.

The digest pulls from your connected sources (Omniscio activity, email, calendar, chats) — sources you haven’t connected simply don’t contribute, so you can turn the briefing on before everything is hooked up. Re-running the mission only updates the hour; your source settings are never touched.

The chat interface is where you communicate with your AI agent. It works like a messaging app — your messages appear on one side, and the agent's responses on the other.

7

Sending Messages

• Type your message in the input area at the bottom
• Press Enter to send
• Press Shift + Enter for a new line without sending
• Press Ctrl + Enter as an alternative way to submit

Paste from anywhere. When a session is open, Ctrl + V (or Cmd + V on macOS) lands in this composer even if your focus is on the sidebar, the toolbar, or anywhere else — you don't have to click into the textarea first. The redirect is skipped when a modal is open, when your focus is already in some other text field, or when the session is ended / archived / paused.

List auto-continuation

When you're typing a bulleted (- , * , + ) or numbered (1., 2)) list in the composer and press the newline key, Omniscio pre-seeds the next line with the next marker so the list keeps going — numbered lists auto-increment (1. → 2. → 3.), bullets repeat, leading indentation is preserved. Press the newline key on an empty list item (marker present, no content after it) to exit the list. With the default submit key (Enter), the newline gesture is Shift + Enter; with Ctrl+Enter submit mode, it's plain Enter. Toggle the behavior at Settings → Sessions → List auto-continuation (default on).

Slash-command menu for skills

When you type / at the start of a message, Omniscio opens an autocomplete menu of your installed skills — each row shows the skill's /command and a one-line description, and the list filters as you keep typing. Use ↑ / ↓ to move the highlight, Enter or Tab (or a mouse click) to insert the chosen /<skill>, and Esc to dismiss. It only triggers on a leading slash, so a slash in the middle of a sentence (like and/or or a /path/to/file) is left alone. The menu lists the skills in your ~/.claude/skills folder plus any installed plugin skills (a plugin skill inserts as /<plugin>:<skill>). Built-in commands like /clear aren't in the menu — type those yourself. Toggle the menu at Settings → Sessions → Slash-command menu for skills (default on).

Fast typing in large sessions

On a very large session (thousands of messages), typing into the message box can start to feel laggy — characters appear a beat after you press the key. Omniscio isolates the message box so a keystroke only updates the box itself instead of re-drawing the whole chat panel, which keeps typing smooth no matter how big the conversation is (the lag is most noticeable in development builds). This is on by default; you can turn it off at Settings → Sessions → "Fast typing (isolated composer)". It's a live toggle — flip it off at any time, with no restart, to return to the classic composer. Drafts, paste, quick replies, voice, and scheduled send all work exactly the same either way.

Message Types

You'll see different types of messages in the chat:

• Your messages — What you typed or dictated
• Agent messages — The AI's responses, including code blocks, explanations, and tool usage
• System messages — Status updates like "Session ready," "Rate limit cleared," or "Conversation compacted"

By default, the chat panel shows just your conversation with the agent and folds away the "plumbing" rows Omniscio generates around it — plain system status events, the "Please continue" prompts Omniscio sends on your behalf (auto-continue, rate-limit / crash / restart recovery), and away-mode auto-replies. If you need to confirm that plumbing is working for a particular session, turn on Show system messages: open that session's three-dot (overflow) menu → More → Filter Messages → Show system messages. It reveals those rows for that one session only, remembers the choice across app restarts, and changes only what's shown — nothing is stored or deleted differently. Status markers like compaction dividers, snooze markers, auth-retry, and rate-limit notices always stay visible whether or not the toggle is on.

Inside an agent's reply, Omniscio also folds each turn's working steps — the tool calls and "let me check that" notes — into a collapsed activity row, leaving the agent's final answer on its own (click the row to see the steps). Omniscio normally works out where that final answer begins on its own. If you'd like it to be exact, turn on Settings → Features → "Show only the final message" (off by default): each session is then taught to mark the start of its final answer, and Omniscio shows you that answer, tucking the working steps before it into the activity row. It never folds a substantive answer away — if the agent marks a bare question or a one-line sign-off, the answer above it stays visible too. It falls back to the automatic behavior whenever a session doesn't mark its answer — so it can only sharpen the split, never break it. A nested option, "Always hide everything before the marker (even without tool activity)", makes that cut absolute: with it on, Omniscio hides everything before the marker even on a plain answer that used no tools, tucking it all into the collapsible activity row (still one click away, never deleted). Off by default; turn it off to restore the standard behavior. A second nested option, "Show every final message when an agent posts more than one", covers a turn where the agent marks more than one final message (say a report and then a sign-off): with it on, Omniscio shows all of them instead of only the last, folding just the work between them into the activity row. Off by default. A third nested option, "Show marker location (debug)", is a diagnostic view: with it on, Omniscio draws a small labelled [[AMC_FINAL]] divider exactly where the agent placed the marker instead of hiding it, so you can see where agents are putting it — it changes nothing about what folds and appears only on turns that actually used the marker. Off by default.

Hovering a link shows its page title

When an agent's reply contains a normal web link (an http:// or https:// URL), hover the link to see a tooltip with that page's actual title — the article headline, the repo name, the video title — instead of just the bare URL. The link's domain appears instantly and upgrades to the real title a moment later, once Omniscio has fetched it; each link is looked up only once. If a page can't be reached or has no title, the tooltip just shows the domain. This is a desktop convenience — on a phone (no hover) the preview never appears and Omniscio makes no web request for it. The fetch runs safely behind the app (only public http(s) pages, never your own machine or private network). The first link of a reply still also shows its Open with Ctrl+Shift+Q keyboard hint.

Re-run from a message (Edit & re-run)

Want a do-over from an earlier point — to reword a message you already sent, or take the conversation a different direction after an AI reply? Hover any message and pick "Edit & re-run from here" (on one of your own messages) or "Re-run from here" (on an AI reply) from its three-dot menu. A small box opens — pre-filled with your message so you can reword it, or empty so you can type the next one — and Ctrl + Enter starts a brand-new session that already knows the conversation up to that point. You land in the new chat with a "Spawned by" link back to the original, which is left completely untouched.

The new session shows the whole earlier conversation with your edited message in place, so it reads like the same chat continued rather than a blank new session. (If the conversation is very long, the assistant only keeps the most recent part in its memory — a small note says so — but you still see all of it above.) It's the way to reword a prompt or explore a different direction without losing your earlier work. The new session runs on the same engine and model the conversation was already using, so you get the same assistant with your correction applied.

Copying a message

Hover any message and open its three-dot menu to copy it. Copy Markdown and Copy Formatted copy the agent's final answer — the same clean reply you see in the bubble, without the tool calls and “let me check that” working notes. On an agent message you also get Copy everything, which opens into As Markdown and As Formatted and copies the whole transcript for that turn instead: every tool call and result, the agent's intermediate narration, and the final answer — the same working steps Omniscio folds into the collapsed activity row. Reach for Copy everything when you want the full trail (to paste into a document, a bug report, or a note); use Copy Markdown / Copy Formatted when you just want the answer. In both, Formatted keeps headings, bold, lists, and code intact for pasting into Word, email, or Google Docs, while Markdown is plain text you can paste anywhere. (One caveat: an agent's private “extended thinking” is never stored, so “everything” means the visible working steps and narration, not the raw hidden reasoning.)

Copying a table. A table inside a message has a three-dot menu at its top-right (hover the table to see it on a computer; on a phone it is always there) with two ways to copy: Copy as Markdown, or Copy table for Excel / Google Sheets. Pick the spreadsheet one when you want the table to land in Google Sheets or Excel as real cells — a markdown paste drops the whole thing into a single cell, because a spreadsheet only splits a paste that carries an actual table. If the rich copy is blocked, it falls back to a tab-separated version that also fills cells.

Code Blocks

When the AI shares code, it appears in nicely formatted code blocks with syntax highlighting. You can copy any code block with one click.

Quick Replies

If you find yourself typing the same responses often (like "yes, go ahead" or "run the tests"), you can save them as Quick Replies. These are reusable snippets you can insert with one click via Alt + S, pin to an Away Mode rule, or fire from an Automation. See Quick Replies.

To manage your library, open the Quick Replies virtual hub in the sidebar.

Each quick reply can optionally be pinned to a specific engine and model — so a canned task always runs on the setup you want instead of the session default. In the reply's editor, open the Behavior tab → Advanced options and flip on Run on a specific engine & model (off by default, so the reply just uses whatever the session is set to). It shows the exact same engine, model, and thinking pickers you use when starting a new session. The pin takes effect when the reply starts a brand-new session; in a chat that's already going it sends on that chat's current engine.

The lightning-bolt (⚡) button in the message toolbar is your Zap — one click (or Alt + Z) instantly sends a single preset reply, the "yes, keep going" button for when the agent just needs a go-ahead (right-click it to change the preset). On a blank session — one you haven't talked to yet — it now asks "Are you sure?" first: the bolt turns amber, and a second click actually sends. That stops an accidental click from firing your canned reply as a brand-new session's opening message, when there's nothing to reply to yet. It only does this on blank sessions — in a conversation that's already going, one click still sends right away. Don't want it? Turn it off at the Quick Replies sidebar entry → its Settings tab → "Confirm on a blank session" (it's on by default).

Not sure what to save? The Eligible phrases tab there lists every phrase you type a lot (skipping ones you can already send in one tap, like your Zap reply) — save any as a quick reply in one click, or dismiss the ones you don't want. It's the browse-all companion to the inbox nudge that offers your most-repeated phrases automatically.

The reverse happens automatically too: when a saved quick reply goes unused for 30+ days, Omniscio drops a card in your Inbox offering to remove it in one tap — with an Undo (a button on the toast, and Ctrl + Z) that restores it exactly. It never touches your built-in replies and never nags about the same one twice. Turn it off at Settings → Features → "Suggest removing quick replies you no longer use".

Markdown Support

Agent responses render full Markdown, including:

• Headers, bold, italic
• Ordered and unordered lists
• Code blocks with syntax highlighting
• Links and images
• Tables
• Mermaid diagrams (flowcharts, sequence diagrams, etc.)
• LaTeX math expressions

Scroll Behavior

While the AI is actively working (green status), the chat automatically scrolls to show the latest output. If you scroll up to read earlier messages, Omniscio respects that and stops auto-scrolling until you scroll back to the bottom.

When you reopen a finished session, Omniscio lands you on the last real reply — it skips past the agent’s automated “waiting on the build…” status notes so you don’t have to scroll down to find the real answer. It also never scrolls above your most recent reply, and counts a reply your auto-reply rules (or Inbox Pilot) sent on your behalf as your reply.

Two small navigation arrows sit at the bottom-right while you’re reading a thread. The down arrow jumps you to the latest message — it shows whenever there’s more conversation below where you’re looking (sized so it still appears on a short phone screen). The up arrow steps back one message per tap, so you can walk up a long thread tap by tap. To jump all the way to the top of the thread, press and hold the up arrow on a phone, or Shift- (or Ctrl-) click it on desktop.

Activity Widget (Loads on Click)

Each agent reply is one bordered card: a header row on top — chevron, bot icon, and a short activity summary like "5 comments · 5 tool calls" — with the agent's actual answer beneath it. The summary counts the work the AI did that turn (files read, commands run, web searches, and the connective notes between them), and it stays visible even while collapsed so you can tell at a glance whether a turn did enough work to be worth opening. The answer itself is always visible — you never have to expand anything to see what the agent said.

Clicking the header expands it and loads that turn's tool detail on demand. Closing it releases the detail from memory. This is what makes long sessions feel instant even when they contain thousands of tool-heavy turns: Omniscio only pays for the activity bytes for the turn you actually want to look at, when you click.

While the agent is actively working on the current turn, the header is auto-expanded so you can watch the tool work as it happens. The instant the turn settles it auto-collapses back to the header row, keeping the activity-summary count on view.

A live timer sits next to the "streaming…" label so you can see how long the agent has been working on the current turn. It counts up once a second — 5s under a minute, 1m 45s up to ten minutes, 12m past ten minutes, 1h 2m past an hour — and keeps ticking even if you expand the bubble. When the turn finishes, the timer freezes into a took 1m 30s badge shown in the turn's expanded view (the collapsed header keeps the activity count, not the duration), so opening any past turn tells you how long that answer took to produce.

Tap an item inside the expanded view (a file read, a search result, etc.) to peek at the full content. Clicking outside or pressing the bubble again collapses it back to the summary.

Opening Long Sessions

Sessions of any size open in a few hundred milliseconds — even threads with thousands of messages. The chat shows the cleaned-up reply text immediately, with each turn's activity collapsed to its header summary; the heavy tool-detail bytes load lazily when you expand a turn (see above). This is true on desktop and on the mobile / phone web view.

You can attach files to your messages so the AI can see and work with them.

Supported File Types

How to Attach

• Click the attachment button (paperclip icon) in the chat input area and select files
• Capture a screenshot of Omniscio (camera icon, next to the paperclip) — snaps a picture of the current Omniscio window and attaches it, so you can show the AI exactly what's on your screen. Off by default — turn the button on at Settings → Sessions → "Screenshot button". Desktop app only.
• Drag and drop files anywhere onto the conversation window — the whole panel becomes the drop target
• Paste images from your clipboard with Ctrl + V
• Paste text — long pasted text (over 5,000 characters) automatically becomes a "chip" attachment

Convert a PDF to Markdown — once a PDF is attached, click the small convert button on its chip to extract its text on-device (free, private, no upload) and swap the PDF for a Markdown text chip that sends inline with your message. It's text-only: a scanned / image-only PDF yields little (there's no OCR).

The File Converter tool

File Converter is a standalone, on-device converter for any supported format — documents (PDF ↔ Word / Markdown and more), images (including HEIC iPhone photos), spreadsheets, audio, and video — free and private, with nothing uploaded. It is in development and off by default: turn it on at Settings → Features, then open it from the toolbar to drag or pick a file, choose a target format, and convert. It works on desktop and on your phone.

It also has an optional Download from web mode — a separate switch you turn on in addition to the converter (Settings → Features, desktop only, off by default). With it off, the File Converter stays conversion-only; with it on, you can paste a link (YouTube, Vimeo, and similar) and save the video or audio to your Downloads. Only download media you have the right to.

Viewing Sent Images

Click any image you've attached to a message to open it. On desktop it opens in a centered viewer card sized to the image — the filename, a position counter, and Copy / Close buttons sit in the card's header, and your workspace stays visible behind a dimmed backdrop. The image opens fitted, never blown up: a small image shows at its actual size, and even a tall full-page screenshot shows whole (capped at about 80% of the window). From there you can scroll to zoom in (or double-click) and drag to pan around it. On a phone the viewer is full-screen with pinch-to-zoom. Use the arrow keys (or the on-screen arrows) to step through several images in the same message, and Esc or a click outside the card to close.

You can turn any AI session into a read-only web page that anyone can open in a browser — no account, no Omniscio install. It's great for showing a colleague what an agent did, sharing an interesting conversation, or keeping a public record of a piece of work.

The shared page shows the AI's written answers in a clean, readable layout — the behind-the-scenes tool steps (files it read, commands it ran) are left out, so the page reads like a finished conversation rather than a work log.

What You Can Share

• A whole conversation — right-click the session and choose Share (the dialog defaults to the full conversation).
• A single message — right-click one message bubble and share just that.
• A highlighted selection — select some text, right-click, and share only what you picked.

How to Share

1. Right-click the session (or a message, or a selection) and choose Share.
2. Pick how long the link should stay live, plus any options you want (below).
3. Omniscio publishes a snapshot and copies the link to your clipboard.
4. Anyone with the link can read it in their browser.

Link Options

• Expiration — every public share carries an expiry: 1 hour, 24 hours, 7 days, or 30 days.
• Password — require a password to open the page. It's hashed on your machine before it ever leaves Omniscio — the plaintext is never sent or stored.
• View cap — limit how many times the link can be opened.
• Notify on view — get a desktop notification the first time someone opens it.

Managing Your Shares

Open Shares in the left sidebar to see every link you've published, each with its view count, last-viewed time, and a "Protected" badge for password-gated links. From there you can revoke a link instantly (it starts returning a "revoked" page), edit its label / expiry / password / cap, or delete it.

Smart Share Reaper (optional, off by default) keeps that list from filling up with forgotten "never expires" links: it finds shares and screen recordings that are over a year old and have never been viewed, gives each a 90-day grace expiry rather than deleting anything outright, and drops a single inbox card asking you to Keep it or let it expire — so an old link you still want is one click from permanent again. Turn it on under Settings → Sharing.

Custom share URL name (paid, Pro+) lets you replace the random token in a share link with a memorable name — shares.omniscio.com/s/my-report instead of a long hex id — and optionally mints a short omnisc.io/<code> link for the same page. If the name you pick is already taken, the share simply keeps its random link. Turn it on under Settings → Sharing.

Sometimes you want to share something an AI agent produced — a chart, a screenshot, a report, a snippet of HTML — without sharing the whole conversation around it. Share Files lets you turn any single file in your project (or any pasted content) into a public URL that anyone can open in a browser.

Unlike Session Sharing, which publishes the whole conversation transcript, Share Files publishes just one artifact. The link points at the file or pasted content by itself, with no surrounding chat history.

Three Ways to Share

From the file peek

When you click a file path in a chat message, the file opens in a peek panel on the right. In the panel's toolbar there's a Share button (the share icon). Click it to publish that file and copy the link to your clipboard.

Right-click any file link in chat

When Claude writes a file path or attaches a file in a chat message — "see docs/notes.md", "the report.pdf you uploaded earlier", anything that renders as a clickable link — you can right-click the link to open a small menu. Below the existing Copy path item is a new Share as public artifact item. Pick it and Omniscio publishes the file straight to a public URL and copies it to your clipboard, exactly the same as the peek-panel button — no confirmation step, no extra clicks. Works on file paths and on attachments you previously dropped or pasted into the conversation. Files that live outside the usual project / ~/Claude / ~/.amc folders — e.g. a file path Claude produced that happens to point at your Desktop — are copied into Omniscio's staging area automatically before being published, so you don't have to move them yourself. (If the link points at something Omniscio can't resolve to a real file on disk — an Inbox or Calendar virtual hub — you'll still see a friendly error toast explaining why.)

From the paste modal

For content you haven't saved as a file yet — a chunk of HTML, a Markdown snippet, plain text, or a React component — open Shares in the sidebar (under the Omniscio group, alphabetically between Settings and Skills) and click "Publish pasted content." Paste your text, pick a content type (HTML, Markdown, plain text, or React), optionally give it a Title (the page's big heading) and Subtitle, and choose how long the link should stay live. For HTML and React you also get a Full-screen toggle plus a backdrop-colour box — see Link Options below. Click Publish and the URL appears, ready to copy.

What You Can Share

• Images — PNG, JPG, GIF, WebP, SVG
• Documents — PDF
• Web pages — HTML files (rendered as a real, interactive web page when opened — scripts run)
• React components — paste JSX/TSX and the page mounts your component live in the browser (React 18 + Babel transpile inline)
• Markdown — rendered into a styled page
• Code — JavaScript, TypeScript, Python, and most common source files (syntax-highlighted, no execution)
• Plain text — any text file

There are size limits per file type (22 MB for images, 18 MB for PDFs, 25 MB for rendered HTML) and the file's extension must match its actual contents — a fake-extension PNG can't sneak through.

Link Options

• Title — sets the page's big heading, the browser-tab title, and the title shown in link previews. Leave it blank and the heading falls back to the filename. The Subtitle (the field formerly called "Label") is the smaller line shown underneath, and is also the friendly name in your list of shares
• Full-screen (HTML, React, and image shares) — drops Omniscio's frame (the centred column, the heading, and the footer) so the page fills the whole screen edge-to-edge. Great for dashboards, slides, and full-bleed designs. HTML and React shares are full-screen by default — the page you publish IS the whole page, with no filename heading or frame; uncheck Full-screen (or pass fullBleed:false from a script) if you want the framed reading view instead. Other kinds (Markdown, text, code, PDF) ignore it and keep the readable column, and an image stays framed unless you turn Full-screen on. On a phone these always show whole, never cut off: a wide desktop-style design is automatically scaled down to fit and centred (pinch-to-zoom for detail), so nothing gets stranded off the side — you don't have to make the design mobile-friendly yourself
• Background (only with Full-screen) — the colour behind the content. Leave it blank and Omniscio picks one for you so it never looks jarring: an image gets a soft blurred version of itself behind it, and a web page gets a neutral that matches light/dark mode. Or type an exact colour (e.g. #0b0e14)
• Expiration — pick from "Never," 1 hour, 1 day, 7 days, or 30 days. After the deadline the link returns a "this share has expired" page
• Password — require a password to view the share. Set it inline from the detail pane and visitors get a password prompt before the content loads. The password is hashed in your machine before it ever leaves Omniscio — the plaintext is never sent across the wire or stored anywhere. You can change it or remove it at any time
• View cap — limit how many times the link can be opened. Once the count hits the cap, the link returns a "view cap reached" page. The detail pane also shows the current view count so you can pick a sensible cap
• Notify on view — turn on per-share to get a desktop notification ("Share viewed: <label>") whenever someone opens the link. Useful for "I want to know when this person reads it" workflows
• Revoke — you can invalidate a link at any time from the Shares view in the sidebar. The URL keeps returning the "revoked" page until you also delete it
• Delete — permanently removes the share and the underlying file from storage

Rich Link Previews (Slack, Discord, iMessage, Twitter)

Every share automatically gets a polished link preview when you paste the URL into Slack, Discord, iMessage, Twitter/X, Teams, or any other app that reads OpenGraph metadata. The preview shows a 1200×630 thumbnail of the share's actual contents, the share's label as the title, and a short description that tells the recipient what kind of artifact they're about to open.

• Images — the preview shows the image itself, scaled to fit the unfurl card
• PDFs — the first page rendered as a thumbnail (so the recipient can see "oh, that's the contract" without clicking)
• HTML, Markdown, code, plain text — an Omniscio-branded card showing the share's label and a small Open in Omniscio… ribbon, so it still looks intentional rather than a raw URL

You don't have to do anything to opt in — the preview is built once at publish time and lives alongside the share. If preview generation fails (network blip, unsupported renderer), the share still publishes normally and the link just unfurls without a thumbnail. Republishing a share regenerates the preview.

Where to Find Your Shares

Open the Shares entry in the left sidebar (under the Omniscio group, alphabetically between Settings and Skills) and Omniscio opens a two-pane layout: a narrow shares sidebar on the left listing every link you've published, and a wide detail pane on the right that shows the currently selected share — with a live preview of the published page and inline controls to edit, send, protect, revoke, delete, or republish it.

The shares sidebar (left). Header shows "Shares" with a count chip and a Publish pasted content button. Below that is a search input that matches the share's label, the URL, and any date strings (so you can type "2026-05" to see shares created or expiring in May). Then a row of segmented Kind tabs (All / Threads / Artifacts / Digests). The list itself is split into two collapsible sections: Live (links still active) and Archived & expired (revoked or expired links). The Archived & expired section starts collapsed so dead links stay out of your way — click its header to expand it. (When you have no dead links at all, the section headers disappear and it's just a flat list.) Each row keeps the label on its own line so it's always readable, with a small lock icon for password-gated links; a muted second line shows kind · scope · time-until-expiry (plus an eye-icon view counter, or N / cap, when you've set a cap), and archived rows carry a small Revoked or Expired badge. Click a row to load it in the detail pane. Middle-click a row to archive (delete) it in place. The list auto-refreshes when shares change in another window or via the CLI.

The detail pane (right). Top of the pane shows the share's label, a status pill (Active green / Revoked red / Expired amber / View cap reached amber), a View button, a secondary Open in browser icon button, and a Copy URL button (with a Copy → Copied confirmation). The pane has three modes:

• Edit mode (default) — shows all the editing controls: label, expiration, password, send, comments, and destructive actions.
• View mode — click View in the header to switch. The edit form is replaced by a full-height in-app viewer that renders the published page inside Omniscio itself — the same content a recipient would see — so you can verify the share without opening a browser. Interactive HTML pages, animations, and tabs all work because the viewer runs in a separate Electron renderer process (not a sandboxed iframe). A floating toolbar appears at the top on hover with buttons for Edit (return to editing), Maximize (go fullscreen), Open in browser, and Copy URL. The toolbar fades out after a few seconds so it stays out of the way. If the page fails to load or the renderer process crashes, you see a clear error overlay with a fallback button.
• Fullscreen mode — click Maximize in the floating toolbar to expand the viewer to cover the entire Omniscio window. Press Escape or click Exit full screen in the toolbar to return to view mode.

Share links clicked inside session messages or anywhere else in the main Omniscio window are automatically intercepted and opened in the in-app viewer instead of launching an external browser, so you never have to leave Omniscio to view a share you already own. In a popped-out session or project window, a share link opens in your system browser instead, because the in-app viewer lives in the main window.

Below the preview is an inline edit form with everything that used to live in a modal: rename the Label, change the Expiration (or check Never expires), toggle a View cap and set the number, turn on Notify on view, and set or change a Password. The password sub-form hashes the value on your machine with PBKDF2 600k iterations before it ever leaves Omniscio — the plaintext is never sent across the wire or stored anywhere. Save is disabled until the form is both dirty and valid; Cancel reverts every field back to what's currently published.

A collapsible Send to… sub-panel sits inline in the detail pane — expand it to send the link to Gmail / SMS / Slack using a 3-tile channel picker. Enter the recipient (email address, phone number in +15551234567 form, or a raw Slack channel ID like C0123ABCDEF), add an optional note (Omniscio appends the share URL after a blank line), and hit Send. You'll see a success toast like "Share sent via Gmail," or a failure toast naming the channel's error if the send didn't go through (e.g. channel_not_found from Slack). Revoked, expired, or unpublished shares fail the pre-flight check before any channel call. Every send — success or failure — writes an audit row to share_send_log tied to that share, so there's always a record of who you sent it to and when.

At the bottom of the detail pane is a destructive-action footer: Update (refreshes a live file-backed share's content in place, keeping the same link — only while active), Revoke (immediate 404 — only while active), Delete (permanently removes the row, with a red-text confirm), and Republish (mints a fresh URL and revokes the old one — only shown when the current share is revoked or expired and has a backing file).

Drag & drop to publish. On the desktop app, drag one or more files from Explorer / Finder anywhere onto the shares sidebar and you'll see a dashed accent border with a "Drop files to publish" overlay. Release and Omniscio publishes each file as its own share, then refreshes the list. Files from your Desktop, Downloads, or anywhere else outside the usual project folders are copied into Omniscio's staging area automatically — no need to move them by hand first. Mixed batches surface a single warning toast like "Published 3, 1 failed: foo.docx — …" instead of one toast per file (typical failures are unsupported file types like Office docs, not path errors).

The Settings → Sharing page is no longer the list view — it now just contains the Firebase Configuration that backs the share publisher.

When you publish a share, viewers can leave comments directly on the share page — no separate tool needed. Comments appear in real time for all viewers and are mirrored back to your desktop app so you can see feedback without opening the share page.

Four Comment Types

• General comments — free-form feedback not attached to any specific part of the content.
• Inline text-selection comments — on markdown, plain-text, code, and SVG shares, the viewer selects text and clicks the "Comment" tooltip that appears. The comment anchors to that selection.
• Coordinate-pin comments — on image and PDF shares, the viewer clicks a point on the content. A numbered pin drops at that spot and the comment anchors to those coordinates.
• Iframe-bridged comments — for HTML and React shares that run inside a sandboxed iframe, the share page bridges text selections out of the iframe so inline comments work the same way.

On the Share Page (Viewer Side)

1. Open the comment sidebar. A floating comment button appears in the bottom-right corner of every share that has commenting enabled. Click it to open the sidebar.
2. Sign in. Google sign-in is required to leave a comment. Click "Sign in with Google" in the sidebar.
3. Leave a comment. Type in the sidebar’s comment box and submit. For inline comments, select text first (or click a point on an image/PDF) and use the tooltip that appears.
4. Real-time updates. New comments from other viewers appear instantly — no refresh needed.

On the Desktop App (Author Side)

• Comments section. Select a share in the Shares view. A collapsible Comments section in the detail pane shows all comments with an unresolved count.
• Allow comments toggle. Inside the share’s edit form, an "Allow comments" checkbox enables or disables commenting per share.
• Inbox notifications. When a new comment arrives, an inbox item is created (controlled by the shareCommentsNotifyInbox setting). The notification shows the commenter’s name for single comments and includes a direct link to the share page.

Settings

• Share Comments Default On (shareCommentsDefaultOn) — new shares have commenting enabled by default.
• Notify in Inbox (shareCommentsNotifyInbox) — create an inbox notification when a new comment is posted.
• Cross-post to Team Chat (shareCommentsCrosspostTeamChat) — when a share is linked to a Team Chat channel, new comments are automatically cross-posted there.

Team Chat Integration

When publishing a share, you can optionally link a Team Chat channel. When a viewer comments on that share, the comment is automatically cross-posted as a thread reply in the linked channel — so team members see feedback without opening the share page.

Managing Channels (Team Chat)

Right-click any channel in the Team Chat sidebar to manage it. The context menu offers Edit Channel (change the name or topic), Leave Channel (for public channels), and Delete Channel (permanently removes the channel and all its messages — you’ll be asked to confirm). Edit and delete are available to the channel’s creator or a workspace admin.

Channel Invitations (Team Chat)

When you create a private channel, you can invite specific workspace members and give each one a role — member (read and post) or read-only (read but not post). Channel admins can also control who can post in a channel: everyone, or managers only. Manage both from the New Channel and Edit Channel dialogs.

Custom Emoji (Team Chat)

Admins can upload custom emoji (party-parrot style — PNG, GIF, or WebP images up to 256 KB) that everyone in the workspace can use in messages, reactions, and custom status. Each custom emoji gets a unique :shortcode: — type it in a message and it renders as the image. Find them in the emoji picker’s Custom tab, or manage them from the admin-only custom emoji dialog.

Link Previews (Team Chat)

When you paste a URL into a Team Chat message, Omniscio automatically fetches the page and shows a rich preview card below your message — with the page title, a short description, an image, and the site name. Up to three links per message get previews; direct image URLs (like .png or .jpg) are skipped because they already show inline. All fetches are SSRF-protected so private or internal network addresses are never reached.

Self-Profile & Availability (Team Chat)

At the top of the Team Chat sidebar, your profile strip shows your avatar, online status, name, and custom status. Click it to open a popover where you can set your availability to Online, Away, Do Not Disturb, or Offline. This overrides automatic presence detection, which is useful since Omniscio runs around the clock. Clear the override to return to automatic mode. The popover also has an Edit Profile button to update your display name and avatar.

You can copy a shareable link to a specific Omniscio page — the Inbox, the Whiteboard, Appearance settings, and more — and send it to a teammate. When they open it, Omniscio opens directly to that page on their own install.

This is different from Session Sharing and Share Files. Those publish actual content (a conversation, a file) as a public web page. A page link is navigation-only — it just tells the recipient's Omniscio which page to open. It contains no private data and is safe to paste anywhere.

How to Copy a Page Link

While you are on a shareable page, look for the "Copy link to this page" control. Clicking it copies the link to your clipboard, ready to paste into a message, email, or Team Chat.

What Happens When a Teammate Opens the Link

• Has Omniscio installed: the link opens Omniscio directly to that page.
• No Omniscio yet: a simple landing page appears with an "Open in Omniscio" button and a link to download the app.

Either way, the recipient sees their own Omniscio — their own sessions, their own data — on the page you linked. The link carries no information from your account.

The Inbox is one of Omniscio's most powerful features. It collects every item across all your sessions and integrations that needs your human attention. Think of it like an email inbox, but for AI agent requests.

9

What Appears in the Inbox?

• Sessions needing attention — Any session with status "Needs You" (the AI asked a question or needs approval). (Updated 2026-08-11: sessions that merely errored or stalled no longer appear here — they go quiet in the sidebar's Interrupted group. Only a session that truly couldn't recover shows up, as "Recovery failed.")
• Cron job approvals — Scheduled tasks waiting for your approval before first run
• Automation approvals — Automation rules that need your sign-off
• Recipe approvals — Recipe steps that require human review
• CLI pending requests — Actions requested via the CLI API that need approval
• SMS messages — Incoming text messages (if SMS integration is set up)
• Daily digest — Summary notifications
• Drip items — Queued reading / link / file items released into the inbox on a cadence you control (see Drip)
• Connection requests — Incoming Team Chat connection requests from other people, shown even if you haven't turned Team Chat on yet
• Team Chat messages — A new message that would notify you (a direct message, an @mention, or activity in a channel you haven't muted) also drops a row here — one per conversation — so you can catch up even if you missed the pop-up. Click it to jump into that conversation; it stays until you dismiss it.

Approving several at once: when approval cards pile up, you don't have to approve them one at a time. The Approvals section shows an “Approve all” button that clears the whole group in one click (with a quick confirm), and you can also select several rows and approve just those together. Each one is still checked individually, exactly as if you'd approved them one by one. Two cards are the exception: the one that deletes old branches, and the one that lets a branch through after a safety check refused it. Approving either does something big, so “Approve all” and “Reject all” leave them where they are, and you decide each on its own card.

Approvals in their own window: the Approvals row in the sidebar (under System) lists the same pending agent approvals. Pop it out with Open in new window and clicking an approval opens it right there, ready to approve or deny — handling one moves you to the next. The list keeps itself up to date there too: new requests appear, and ones handled elsewhere drop off.

Approvals inside the conversation: an approval an agent raises also shows up as a card just above that session's message box, so you can answer it without leaving the thread. The card shows at most four lines of the request and two of the agent's reason, so its Approve and Decline buttons always stay on screen, even on a phone; tap Details to read everything. When several cards are waiting at once, they share an area that scrolls, so every button can always be reached. A card asking you to approve a change to outside packages names the branch and how many files it changes; Details lists the files and each package the change adds, removes or updates, with its old and new version, and says when a file also changes other install settings.

Switching to Inbox View

Click the "Inbox" tab at the top of the left sidebar. The badge number shows how many items need attention. Inside the list, each project — and each source like Approvals or SMS — also shows its own count in parentheses next to its name, so you can see at a glance where your attention is needed.

Working Through the Inbox

You don't have to click each item to begin — the Inbox keeps one open for you. Finish the current item and it drops you straight into the next, and if a session starts needing you while you're looking at the empty "All clear" screen, it opens on its own. You're always looking at the thing that needs you, never a stale empty pane. The list scrolls to follow your selection, too — clear an item and the next one is brought into view, even if it sits above where you'd scrolled, so it never ends up hidden off-screen.

And if a session you're reading comes back to life on its own — it was waiting on you, then a background check-in starts it working again — you stay on it and watch it run, instead of being bounced back to "All clear." Anything new that needs you waits in the sidebar with its badge until you're ready for it.

Approving with the Keyboard

When an approval is open — a settings change, a session pause or archive, a cron job, a recipe, and so on — you can act without the mouse: press Enter to approve and X to reject. A faint reminder sits in the corner of the approval. Typing in a text field won't trigger them, and holding a key won't blast through your whole queue — each press counts once. (These two keys are fixed, not rebindable.)

Settings-change approvals read in plain English, too: instead of Set quickLaunchTabHotkeys: {"calendar":"CommandOrControl+Alt+J"}, you'll see "Set Quick Launch shortcuts (Calendar) to Ctrl+Alt+J" — the raw details are still one click away under Show technical details.

Multi-Select

You can select multiple inbox items at once to batch-process them:

• Shift + click to select a range
• Shift + J / Shift + K to extend the selection
• On mobile, long-press to start multi-select mode
• The session or item you already have open is included automatically — tap (mobile) or click (desktop) it to leave it out

Once multiple items are selected, you can act on them all at once. Archive them together, or right-click the selection for the full set of bulk actions — Pause / Unpause, Stop, Restart, Snooze, Pin / Unpin, Move to another project, Set color, and Copy link — each applied to the sessions you selected, with a single Undo where the action supports it.

Archiving Inbox Items

• Desktop: Middle-click (mouse wheel click) an item to archive it
• Mobile: Swipe or use the close button

Snoozing Inbox Items

Every row in the Inbox is snoozable — sessions, SMS conversations, Telegram chats, fired alarms, cron / automation / recipe / recipe-authoring / CLI Pending approvals, RSS articles and digests, drip items, the daily digest, and weekly summaries. Right-click any row to see a Snooze submenu with preset durations (2 hours, this evening, tomorrow morning, etc.) or pick Custom... to type a natural-language time like in 2 hours, friday 9am, or compact shorthand like 805p — the same Snooze Palette and the same smart AM/PM defaulting as session snooze. The row disappears from the Inbox immediately and reappears automatically when the snooze expires (checked every 60 seconds). A fired-alarm row also keeps its inline Snooze 9m button for one-tap dismissal.

Notification History

Toasts — the small pop-up notifications in the corner — disappear after a few seconds, and on a phone they now stack behind each other as a compact deck (newest in front) so a burst never fills the screen. To look back at ones you missed, open Notification History from the System group in the sidebar — it works the same on desktop and mobile. It has two tabs: Toasts, a saved list of recent pop-ups that survives restarting the app, and Alerts, the full history of agent alerts you've dismissed (beyond the Alerts screen's last-7-days view) — each one can be Restored back into your current alerts.

Who Triggered This? (Provenance)

When an AI agent makes a request through Omniscio's CLI control API — creating an approval or spawning a new session — Omniscio records which session caused it, so you can trace it back:

• "Generated by <session>" — shown in an inbox item's detail view (open the approval, or open an agent-raised alert) for agent/CLI-created items. On a pending approval it also shows the exact date and time it was generated. Click the session name to jump straight to the agent that caused it.
• "Spawned by <session>" — when one agent spawns another, the new session opens with a pinned note at the top linking back to its parent. Follow the chain to see who spawned what.
• "via CLI · from <session>" — when an agent sends a turn into a session over the CLI (a message, a nudge, a peer message, or a scheduled reply), that turn's bubble carries a small tag linking back to the sender — so an injected message reads as from that agent instead of an anonymous "You" turn.
• "injected by <sender>": when Omniscio itself sends a note into an agent's session (your Inbox approve or decline, an Auto-lander or Gate watch result, a scheduled wake-up), the transcript shows it as a short card that says what happened in plain words. Click Show the original note to see the exact text the agent received; the agent's own copy is never changed.

Agents that Omniscio launched identify themselves automatically. A request you make by hand (or an external script) simply shows no origin — provenance is a tracing breadcrumb, not a security check.

Away Mode automatically responds to your AI agents when they ask questions. It's like an out-of-office auto-reply, but for your coding sessions. When enabled, Omniscio will answer common agent questions on your behalf so work keeps moving.

How to Enable

1. Open the Automations entry in the sidebar and switch to the Auto-replies view — the Away Mode rule set lives there, not in Settings.
2. Toggle "Away Mode" on.
3. Set your default auto-response message (e.g., "I'm away. Use your best judgment and continue.").
4. Optionally, create specific rules for different scenarios.

Creating Rules

Rules let you customize how Omniscio responds based on what the agent is asking. For example:

• "If the agent asks about testing" → respond with "Yes, run all tests"
• "If the agent asks about deploying" → respond with "Do NOT deploy — wait for me"
• "If the agent asks anything else" → respond with the default message

Each rule has:

• A condition — What triggers this rule (keywords, patterns, or "any question")
• An action — What Omniscio does when the rule matches
• Priority — If multiple rules match, the highest-priority one wins

Beyond a plain reply, a rule's action can also organize the session — in one step, give it a new title, apply a tag, and set a custom inbox color (fill in any combination). It can also rename it, archive it, start a new session, or drop a note in your inbox. "Organize the session" is handy for triage: when an agent reports it's done, a rule can retitle it "Ready for review" and tint it green so it stands out at a glance.

A session's color shows as a thin colored bar on its row in the inbox and sidebar — alongside, never replacing, its status dot. You can also set or clear a color by hand: right-click a session and choose Set color (pick a swatch, or No color to clear it).

Auto-replies are loop-safe: if a rule keeps replying to a session that has nothing left to do, Omniscio stops after a few tries and surfaces the session to you with a short note — so a rule left running overnight can't quietly spend turns in a loop. It picks back up as soon as the agent does real work again.

Silence Until

You can set a "Silence Until" time, which mutes all notifications until that time. This works independently of Away Mode — you can silence notifications without enabling auto-responses, or use both together.

The Dev Pipeline turns "build me X" into a disciplined six-phase workflow your agent follows on its own isolated copy of your code. You approve at five checkpoints (called gates), and the run ends with a branch that's verified and marked ready to merge — the pipeline itself never merges or pushes anything — the auto-lander, which the Dev Pipeline arms, is what lands it afterwards.

Dev Pipeline Maintenance — turning the Dev Pipeline on also installs two housekeeping skills that keep your worktrees tidy: worktree-cleanup (reaps git worktrees whose branch has provably merged) and merge-all-ready (lands every local branch marked ready-to-merge, atomically and local-only). Once a day a background check looks over your worktrees and ready branches on the repos you had dev-pipeline activity in — it costs nothing and always leaves a receipt. It only starts those two skills when it finds something it cannot settle on its own, so on a tidy day nothing runs. Controlled from the Dev Pipeline panel's Setup tab (devPipelineDailyMaintenanceEnabled, on by default).

Returned cloud work — when a cloud session ends with changes, its work comes home as an ordinary branch and is registered for landing. With this on, the app lands that branch by itself, so you never have to run ready-to-merge on a returned branch by hand. Which way it does that depends on the project. For this project it tests the branch and marks it ready to merge, and the auto-lander takes it from there. For any other project it runs that project's own test command on the cloud machine — against the very files coming home, on the one machine where that project's dependencies are installed — and merges the result into that project's own main branch here on this computer, once those tests pass. Five things still wait for you: a branch that changes a dependency or that changes a project's own test setup, one whose files could not be checked for secrets, one whose check did not cover its own files, and one that came back with an answer nothing can change — a red check or a merge that will not go cleanly, a test run that could not start or ran out of time, a project with no test command, or a project folder with unsaved changes. The session's own transcript tells you what happened. Controlled from the Dev Pipeline panel's Setup tab (cloudReturnedWorkAutoLand, on by default; it does nothing unless you are using cloud sessions).

Turning It On

1. Go to Settings → Features.
2. Toggle "Enable Dev Pipeline skill" on. This installs the /dev-pipeline command for every project (and reveals the Dev Pipeline panel in your sidebar).
3. In any session, type /dev-pipeline followed by your task.

The Six Phases

1. Plan — the agent investigates and presents a plan. Questions it genuinely needs you to answer appear right at the top.
2. Red team — the plan is stress-tested through ten critical lenses before any code is written (small, low-risk tasks get a fast pass; big ones get full depth).
3. Build — the work happens, test-first, with proof it's green.
4. Polish — a cleanup pass that only makes safe, behavior-preserving tidy-ups; anything bigger is flagged for your call.
5. Docs — documentation is brought in line with what changed.
6. Ready to merge — one final full verification, then the branch is stamped ready. Done.

Standards Checks on Every Run

On top of those six phases, every run also does two quick standards checks that approve themselves unless they find a real problem: a pre-build check right after the red team, and a pre-merge audit right after docs. They keep each change lined up with your project's own engineering standards — reuse before building something new, the right shared components, tests, docs, security, and so on. Both appear in the Dev Pipeline panel's gate list, where you can turn either one's auto-approval off to have the run stop for your sign-off instead.

What they check comes from a plain file you can customize per repo: .claude/dev-pipeline/standards.md, with a "Before you build" section and a "Before you merge" section. Commit that file and it travels to everyone on the project. If a repo doesn't have one, the checks fall back to a sensible general standards file that ships with the skill — so they still help in any repo, and a missing file never blocks a run. To switch the two checks off entirely, set DEV_PIPELINE_DISABLE_BUNDLED_STANDARDS=1.

Gates and Auto-Approval

Each of phases 1–5 ends at a gate where you reply approve, give feedback, or stop the run. You decide this per gate: each of the five gates (Plan, Red Team, Build, Elegance, Docs) has its own switch for whether Omniscio approves it for you or always asks, plus a select-all toggle to flip them together (Settings → Features, or the Dev Pipeline panel). By default the Plan gate asks you while the other four approve themselves the moment a phase reports done — so a healthy run flows hands-free from your one plan approval to the finish line, and you can tighten or loosen any single gate.

Every gate report opens with a traffic-light dot that tells you at a glance whether you're needed: 🟢 green means nothing needs you (safe to just approve), 🟡 yellow means there's a question to answer, and 🔴 red means a serious call or a blocker. When the dot isn't green the title says so too (e.g. “Plan Not Ready — One Question”), and the little pipeline track under the report lights up to match.

Getting Your Finished Work Onto master

A finished run leaves you a branch marked ready to merge — verified, but not yet part of your main branch. “Landing” is the last step that folds it in. Three ways to get there:

1. Let the auto-lander do it (the hands-off option, and the default one). The auto-lander is already on: enabling the Dev Pipeline arms it for every hub the pipeline covers, and it merges every ready branch into that repo’s main branch by itself — locally, on your machine, while Omniscio is open. It only ever moves work forward and never writes to a remote. From the Dev Pipeline panel’s Setup tab you can narrow which repositories it watches, set one to observe-only, or turn it off entirely (the master switch also lives at Settings → Performance → Sessions). Watch it work on the Auto-lander tab.
2. Ask for it once. In any session, say /merge-all-ready. It lands every local branch currently marked ready, one at a time. A branch that can’t land safely is skipped and reported, never forced.
3. Do it yourself. The branch is an ordinary git branch — merge it however you normally would.

You don’t have to choose once and for all: the daily maintenance job described above will start merge-all-ready for you when it spots ready branches piling up in repos you’ve been working in — but it is a backstop, not a schedule. The auto-lander is what keeps the queue short day to day. And when a branch does land, you’ll see it twice — the session that produced it gets a green “Landed … to master” note, and the run is marked “Landed ✓” in the Timing tab.

Approving Changes to the Merge Tooling

While it merges your finished branches, the app runs a few small helper programs that come from your project. Together these are called the merge tooling. They used to run straight from whatever was on disk, so an edit to one of them quietly became code the app ran on the next merge, with nobody checking it first.

Now the app runs that tooling only from a version a person approved. When the tooling changes, the app raises one card in your Inbox asking you to approve the new version. It only appears when the tooling actually changed, so ordinary merges never show it.

The card tells you what you are approving:

• Which project it is, by name.
• How many tooling files changed, and their names.
• Which commits changed them.

It deliberately does not show a long commit id.

Approve and your merges switch to the new version. Decline and your merges keep using the version you last approved, and that exact version is not asked about again. Nothing is lost either way.

The app keeps its own copy of the version you approved, prepared from exactly that version and checked before every merge that uses it. The helper programs are pointed at that copy for that one merge only, so nothing else changes: your own merges, other projects and anything else running at the same time behave exactly as before.

If that copy cannot be prepared, or fails the check that it is still intact, the app does not quietly fall back to running whatever is on disk. The files those helpers would have handled are merged with the app's own built-in merge instead, and if it keeps happening you get a card telling you. There is one switch that puts the old behaviour back, and it is on by default; it is deliberately kept out of Settings, because the thing it turns off is a safety guarantee.

The Dev Pipeline Panel

Enabling the skill also reveals a sidebar panel. Its left rail has six tabs (five always, plus Agent Board when you have the Agent Status Board turned on):

• Live — every running pipeline grouped by project, with the step it's on, an amber "waiting on you" counter when one is parked at a gate (with a one-tap Approve), the run's approximate spend so far, and your git worktrees with guarded one-click cleanup.
• Timing — where recent runs' time and money went: working vs. waiting-on-you vs. stuck, plus an approximate cost per run and per phase, so you can spot bottlenecks and expensive steps.
• Agent Board — every agent Omniscio has launched, each a card with its status, workspace, pipeline step, and live to-do list. A card updates as soon as its agent does something, and the board is back the moment Omniscio restarts. A card also says when an agent started without a to-do tool, or when the to-do rule let it skip keeping a list. Click a card to jump to that session. This tab appears only when Settings → Workflow → "Agent Status Board" is on.
• Auto-lander — the queue and history of work reaching your main branch: what the lander is doing right now, which branches are lined up to land, and every branch it has landed, handed back on a conflict, or set aside as "Couldn't land". You can pause or resume the whole lander, or just one repository, from here.
• QA Fleet — the results of Omniscio's own background QA runs: the open findings it has raised, which you can resolve, dismiss, or reopen (with a one-click undo), per-feature end-to-end pass/fail, the features that have gone longest without being exercised, and a findings trend over time. It refreshes every 30 seconds, and shows an amber count when findings are open.
• Setup — the pipeline's controls in one place: gate auto-approval, the workflow companions, which repos are covered, custom phases, AI code review, Auto-Lander (per-repo config), worktree cleanup, git guardrails, and Repo Foundations. Under AI code review you pick which AIs double-check the work — every step that runs is listed, including any your repo sets for everyone, and you can switch one off for yourself without touching the team's setup. One switch at the top, Run AI code review steps, turns every review step off at once — even in runs already under way — while keeping your choices for when you turn it back on. A scorecard right below keeps score for each AI: how many reviews it did, the serious and minor problems it found that turned out to be real, its false alarms, and what it cost (“not reported” when its company shares no cost, never a made-up $0).

Git guardrails

Git guardrails protect your shared branches from every agent Omniscio runs — not just Dev Pipeline runs. When on, an agent is blocked from committing or pushing to a protected branch (master/main), editing files while a repo sits on one, or running a destructive git stash. It's on by default; you can turn it off, switch it to warn-only, or edit the protected-branch list under Settings → Features → "Git guardrails" or in the Dev Pipeline panel's Setup tab.

Need to do some protected-branch work yourself? From the same controls you can pause the guardrails for a few minutes (with a live countdown) or exclude a specific repo. These controls are yours alone — because pausing or excluding lowers the guardrail, they're deliberately unavailable to agents and to a paired phone, so an agent can never switch off its own guardrail.

Developer guardrails

Developer guardrails are the machine-wide sibling of Git guardrails. Instead of protecting only the sessions Omniscio launches, they install a single safety hook into your global Claude config (~/.claude/settings.json) so that every Claude Code session on this machine — even ones Omniscio never started — is blocked from committing, pushing, or editing on a protected branch (master/main) or running a destructive git stash. They also refuse anything that would put the full-trust CLI key into a session's output — reading or grepping the key file, a shell command that names it, overwriting it, or fetching the same secret from the vault — while the inline form every agent uses, -H "Authorization: Bearer $(cat ~/.amc/cli-token)", keeps working. They reuse the same engine as Git guardrails, including its pause and per-repo exclusion escape hatches. Because turning them on edits your global config, they're off by default (opt-in); enable them under Settings → Features → "Developer guardrails".

When a Run Gets Stuck

Five things can hold a run up. Each looks different, and each has a way out:

• A gate is waiting on you. The run stops at a gate with a 🟡 yellow or 🔴 red dot and an honest title (“Plan Not Ready — One Question”). The Live tab shows an amber waiting counter next to it. Answer the question in the session, or hit the one-tap Approve on the Live tab. Nothing is lost while it waits.
• The build won’t go green. The run opens with a red Build Blocked header and stops rather than faking a pass or looping. Read what it hit, reply with a steer, and it picks up from there.
• The branch conflicts and gets handed back. The auto-lander never force-merges. When a ready branch collides with your main branch it hands the branch back to the session that wrote it — reopening that session even if it was archived or its worktree was cleaned up — to rebase and re-mark it ready. You’ll see it on the Auto-lander tab as a Conflict row, with the exact files it collided on. Usually you do nothing; it comes back around on its own.
• Landing it would drop work someone else already merged. A safety check runs before every land, and when a branch would throw away another change’s work on your main branch, the auto-lander holds it and tells the session that wrote it first, or the session that launched it if that one is still working, and it wakes a session that archived itself when nobody else can take it. That session rebuilds the branch so both sides are kept, which is the usual fix. You only get an inbox card when nobody can take it (you closed or paused the session) or when it still isn’t fixed four hours later. If the session decides the removal really is intended, it signs off on it, and when the branch lands you get one card naming the files and its reason.
• The branch was set aside — “Couldn’t land”. This means the lander tried and stopped on purpose: a safety check flagged something (a possible secret in the diff, or new text that isn’t translated), it ran out of automatic retries, or the work had already reached master another way. The branch and its commits are safe — parked, not lost. Open the row on the Auto-lander tab to see the reason, or start a session from the inbox card and an agent will clear it.

Most of this never reaches your inbox on purpose. Routine churn — a branch simply waiting its turn, or a retry in progress — is recorded on the Auto-lander tab instead of interrupting you. You’re pinged only when something genuinely needs a person: the pile of ready branches grows past its threshold, nothing is landing at all, or a possible secret needs your judgment.

Rescuing a Stranded Run

If a session dies mid-run (a crash, a closed laptop, a deleted session), its work isn't lost — the worktree keeps the run's full saved state. The panel marks it with a grey "Stale run" pill and an Adopt button: one click starts a fresh session in that same worktree that picks the run up exactly where it stopped. Adopting never touches a run whose session is still alive, and it only ever happens when you click.

Hands-Free Mode

Ask for it explicitly — "run autonomously", "leave it overnight" — and the run self-advances through every phase with the same rigor, stopping only for something that genuinely needs you: a real decision, an ambiguous ask, or a build it can't get green. It still ends at a merge-ready branch, never a merge.

Every session that uses a git worktree leaves it on disk when the session ends. Over time these pile up — dozens of folders that served their purpose but are still taking space. The /worktree-cleanup skill looks at every worktree in a project, works out which ones hold work that has already landed, and clears those out while leaving everything else alone.

It works out “already landed” by comparing the actual content of your commits against your main branch — not by branch name, and not just by asking whether the branch was merged directly. That means it still recognises work that landed by squash, rebase, or replay, which a simpler check would miss and leave sitting on your disk forever. It is plain git: GitHub is used as one extra hint when it happens to be available, and is never required.

How to use it

1. In any Claude Code session, type /worktree-cleanup or say “clean up worktrees.”
2. The skill lists every worktree, sets aside the ones that are clearly still in use, and checks the rest against your main branch.
3. It presents a plan with three buckets: Landed (safe to reap), Keep (real work that hasn’t landed yet), and Not sure (kept and reported — never removed on a guess).
4. Tell the AI which to remove — all the landed ones, specific picks, or none.
5. It saves anything unsaved first, notes how to restore each branch it deletes, removes the approved worktrees, and tells you what it did and where the backups are.

What it never touches

• Anything still in use — a worktree that is locked, mid-operation, marked active, or is your main checkout.
• Anything with unsaved work — uncommitted changes are saved to a backup patch first, and nothing real is ever discarded.
• Anything it can’t prove has landed — when it isn’t sure, it keeps the worktree and says so.
• Any branch whose work isn’t in your main branch — the worktree may go, but the branch stays, so the commits are still there.

When it runs without asking

The walkthrough above is the interactive run — you say what to remove and nothing goes without your word. The same skill also runs unattended, and it’s worth knowing where:

• Unattended mode (/worktree-cleanup --auto) removes only the worktrees it can prove have already landed — anything ambiguous is skipped and reported, never guessed at. It never waits for an answer.
• The daily maintenance job can run exactly that unattended mode for you — it checks daily in the repos you had Dev Pipeline activity in and starts the skill only when the check finds a branch needing a person’s judgement. It’s on by default whenever the Dev Pipeline is enabled. Turn it off in the Dev Pipeline panel → Setup tab.

So yes — folders can disappear overnight, on purpose. What protects you is that nothing is ever lost: before a worktree goes, the skill records its branch’s recovery commit and saves a patch of any uncommitted changes, so a removal you didn’t want can be restored. A branch marked do not delete is never touched.

Automations are "if this, then that" rules for your AI sessions. You define a condition (the trigger) and an action (what happens). Omniscio watches your sessions and fires the automation when the condition is met.

In the sidebar, these live as two rows under the Automation group: Automation Rules — the powerful engine described here, with multi-step action chains and AI conditions — and Auto-replies, the lightweight cousin where one simple trigger (a keyword, project, tag, or time) fires a single action. Reach for Auto-replies first; graduate to Automation Rules when you need a full pipeline.

Creating an Automation

1. Go to Settings → Automations (or access via the Automations virtual hub).
2. Click "New Automation."
3. Choose a condition (what triggers it).
4. Choose an action (what happens).
5. Optionally limit it to specific projects.
6. Save — new automations require approval in your Inbox before they become active.

Available Conditions

Available Actions

Automation Runs

Every time an automation fires, Omniscio logs it. You can see the history of all automation runs, including what triggered them and what action was taken.

Saved Credentials

Actions like "send a Slack message" or "post to a Discord webhook" need API tokens and other secrets to call those services. Omniscio stores them centrally in Settings → Automations → Credentials — encrypted on disk using your operating system's keyring (Windows DPAPI, macOS Keychain, or Linux libsecret). Eight kinds ship today: Slack, GitHub, Discord webhook, Notion, Telegram, generic HTTP auth, Google service account, and SMTP. Two of those (Google service account and SMTP) are gated behind an opt-in toggle because a leaked key would expose broad data access or an outbound mail relay; creating or rotating one is blocked until you flip the gate on. Automations reference credentials by id, so you can rotate a secret without editing any rules.

Recipes are Omniscio's most powerful automation feature. A recipe is a sequence of steps that run one after another, where each step is a task for an AI agent. Think of it like a cooking recipe — you define the steps, and Omniscio executes them in order.

How Recipes Work

1. Define steps — Each step has a prompt telling the AI what to do.
2. Run the recipe — Omniscio executes steps in sequence, starting a new AI session for each step.
3. Review & approve — Some steps can require your approval before proceeding.
4. See results — The Run Dashboard shows the status of each step.

Recipe Concepts

• Steps — Individual tasks in the recipe. Each step runs as its own AI session.
• Approval gates — Points where the recipe pauses and waits for your OK before continuing.
• Context passing — Results from earlier steps can be passed to later steps so the AI has context.
• Memory — Recipes can store and recall information across runs.
• Patterns — Pre-built recipe templates for common workflows (see Recipe Patterns).

Omniscio comes with several recipe patterns — ready-to-use templates that define common workflows. You can use them as-is or customize them for your needs.

Available Patterns

Using a Pattern

1. Open the Recipe editor.
2. Click "Use Pattern" or select a pattern from the template gallery.
3. Customize the step prompts for your specific use case.
4. Save and run the recipe.

Connecting Mobbin (design research)

The Design Overhaul recipe can ground its work in real shipped designs by searching Mobbin, a library of screenshots and user flows from thousands of apps. This is off by default and only needs to be set up once. It works with both free and Pro Mobbin accounts — whatever your Mobbin account can see, the recipe can use.

1. Go to Settings → Sessions and find the Mobbin MCP (design research) card.
2. Click "Setup Mobbin MCP". Omniscio registers Mobbin with the Claude Code CLI in the background — no terminal window opens and no session is spawned.
3. When prompted, paste your Mobbin browser cookie: log into mobbin.com, open the browser console (F12, then the Console tab), run copy(document.cookie), then paste it into the box and click Authenticate.
4. On success the card shows "Mobbin MCP configured and authenticated." and the toggle flips on. Run the Design Overhaul recipe and its first step researches Mobbin automatically; if Mobbin is not connected the overhaul still runs, just without the reference brief.

Starting a Run

1. Open the recipe you want to run.
2. Select the target project (which codebase the AI works on).
3. Click "Run Recipe."
4. The recipe starts executing its first step.

The Run Dashboard

Click the Recipes button in the toolbar (or navigate to the Recipes section) to see the Run Dashboard. It shows:

• All active and recent recipe runs
• Each step's status (pending, running, completed, failed, needs approval)
• Timing information for each step
• Error details if a step fails

Approval Steps

Some recipe steps are configured to require your approval. When a recipe hits an approval step:

1. The recipe pauses.
2. An approval request appears in your Inbox.
3. You review what the previous step did.
4. Click "Approve" to continue or "Reject" to stop the recipe.

Silent Recipes

Recipes can run in silent mode, which hides their sessions from the sidebar and inbox. This is useful for background tasks you don't need to monitor constantly. Silent recipe sessions appear in a dedicated section and have a watchdog that surfaces them if they've been running too long (over 6 hours).

Error Recovery

If a recipe step fails, Omniscio can:

• Retry the step — Run the same step again
• Skip and continue — Move to the next step
• Stop the recipe — Halt execution entirely

Cron Jobs let you schedule tasks to run automatically at specific times. It's like setting an alarm clock for your AI agents — "Every morning at 9am, check the codebase for issues."

My Automations

My Automations is the friendly home for everything you've set to run automatically. Open it from the Automation section of the left sidebar to see one list of all your automations — the rich ones built with the Automation Helper (marked with a badge) alongside the simpler jobs you scheduled yourself. Each row shows what it does, when it runs, and its status. It's the one place the app points you to after you approve an automation; everything here also appears under Cron Jobs for the low-level details.

Creating a Cron Job

1. Go to Settings → Features → Enable Cron Jobs.
2. Click "New Cron Job."
3. Set the schedule (e.g., daily at 9am, every hour, weekly on Mondays).
4. Write the prompt — what you want the AI to do each time.
5. Select the target project.
6. Save — the cron job requires approval in your Inbox before its first run.

Schedule Format

You can set schedules using simple options like "Daily," "Hourly," or "Weekly," or use cron expressions for advanced scheduling:

0 9 * * *     = Every day at 9:00 AM
0 */2 * * *   = Every 2 hours
0 9 * * 1     = Every Monday at 9:00 AM
0 9,17 * * *  = Every day at 9:00 AM and 5:00 PM

Silent Mode

Cron jobs can run in silent mode, which means their sessions won't clutter your sidebar or inbox. The results are logged, but the sessions stay hidden unless something goes wrong.

Run if Missed

A cron job only fires when Omniscio is running at the scheduled time. If Omniscio was closed — your laptop was asleep, or you were away for a few days — that scheduled run is normally skipped. Turn on "Run if missed" in a job's editor and Omniscio will instead run that job once the next time you open it, catching up misses from the last 7 days.

It's off by default, so existing jobs are unchanged. However many runs were missed while Omniscio was closed, the job runs exactly once on next open — not once per missed day — and then resumes its normal schedule. The catch-up run still respects approval (if the job requires it) and still costs the same as one normal run. The toggle only appears on recurring jobs, since a one-time job has no schedule to catch up.

Approval System

For safety, every new cron job and every edit to an existing cron job requires your approval before it runs. This prevents accidental automated spending.

Alarms are phone-style reminders for Omniscio. Type something like "tomorrow at 9am" or "every Tuesday at 2pm stand-up" into a single textarea and Omniscio saves the alarm, schedules the fire, and rings it however you've asked — full-screen modal, toast banner, or silent inbox row. When it fires you get Snooze and Dismiss; recurring alarms re-arm themselves automatically, and one-off alarms vanish from the sidebar the instant they fire (phone-alarm semantics) — only the Inbox row remains so you can still see it happened. Snooze and Dismiss resolve the alarm on every device you have open — clear it on your phone and the ring stops on the desktop too, without you touching it.

Alarms are separate from Cron Jobs even though both are time-triggered. Cron jobs run code (a recipe, a session spawn) and post results to the inbox; alarms only ring at you — no command, no working directory, no agent spawned, no cost.

Creating an Alarm

1. Open the Alarms virtual hub in the Omniscio sidebar.
2. Type the alarm in plain English into the textarea at the top ("every weekday at 7:30 AM stand-up").
3. Press Enter to save. The new alarm appears under Today, This Week, Recurring, or Disabled.
4. Refine the label, sound, snooze duration, or per-alarm overrides from the editor on the right.

Quick Add from Header

You don't have to open the Alarms project to schedule a one-off. The toolbar has a bell-plus icon in the header (visible whenever Alarms is enabled) and the keyboard shortcut Ctrl+Shift+A opens a centered modal from anywhere in the app.

The modal has a natural-language textarea at the top — type something like "tomorrow at 9am dentist" and press Enter to parse it into the editable fields below. From there you can adjust the Label, Time, Recurrence, and (when applicable) Date or Days-of-week, then press Ctrl+Enter or click Create alarm to save. New alarms land in Uncategorized; sensible defaults (08:00 daily) apply if you skip the textarea.

Natural-Language Examples That Work

• at 8pm — one-off, today (or tomorrow if 8 PM has already passed)
• 805p / 8p / 1230a — compact shorthand for 8:05 PM / 8:00 PM / 12:30 AM (resolves on the free offline parser, no AI call)
• tomorrow at 9:30am — one-off on a named day
• every day at 7am — recurring, daily
• every weekday at 7:30 AM stand-up — recurring, Mon–Fri, label "stand-up"
• Mon Wed Fri at 6:45 PM yoga — recurring on specific weekdays
• remind me to take meds at 8pm — label phrase preserved alongside the time

Organizing With Folders

Beneath the time-based sections (Today / This Week / Recurring / Disabled) you can create flat user-defined folders to group related alarms together — "Work", "Meds", "Kids", whatever makes sense for you. Click + New folder, type a name, press Enter. Hover a folder header to rename (pencil) or delete (trash) it.

File an alarm into a folder by dragging the row onto the folder header, or by picking the folder from the dropdown in the alarm editor on the right. Deleting a folder doesn't delete its alarms — they move back to Uncategorized. Folders are purely cosmetic: an alarm rings the same way whether or not it's in a folder.

Per-Alarm Sound & Behaviour

Each alarm carries its own sound configuration plus a few behaviour knobs. Open the alarm in the editor (Pane 3) and click Advanced to see them. Every knob can be set per alarm or left as Use default to inherit the global value from Settings → Notifications → Alarms.

• Play mode — once plays the sound one time; loop keeps playing it. Loop applies to both modal alarms and banner toasts.
• Duration — how long the sound is allowed to ring for. Useful with loop mode (5s / 10s / 30s / 1m / 2m / 5m / Forever).
• Fade-in — ramps the volume from silence up to the resolved level over N seconds (0 = off, 2 = quick crescendo, 60 = gentle wake).
• Volume override — per-alarm volume that ignores the global notification slider. 10% / 25% / 50% / 75% / 100%.
• Speak label aloud — Omniscio speaks the alarm's label via the OS voice on top of whatever sound the alarm plays. Handy for accessibility or for hearing an alarm from another room.
• Max snoozes — cap the number of snoozes per ring chain. Unlimited (default) keeps the Snooze button always. Dismiss-only (0) hides Snooze entirely. Pick a number (1/2/3/5/10) and the Snooze button disappears after that many snoozes.
• Require math to dismiss — a per-alarm anti-snooze gate. With this on, you have to solve a small arithmetic problem (e.g. 7 + 8 = ?) before Dismiss becomes clickable. Snooze is unaffected.

Test Fire — Preview an Alarm Without Scheduling It

The alarm editor's footer has a Test fire button next to Delete. Click it and the alarm fires immediately — the modal opens, the sound plays with your chosen fade-in and volume, the label is spoken if you've enabled that, the Snooze/Dismiss buttons render — but the alarm's schedule is untouched. The fire isn't counted, next_fire_at isn't advanced, and no missed-alarm inbox row is created.

Use it to pick a sound, test loop behaviour, verify a fade-in, or check how hard the math gate is before going to bed. The math gate is forced off in previews so you can always close the test fire.

Mobile Inbox

Missed alarms surface in your Inbox so you can see them on your phone even when Omniscio wasn't open. Pierces Focus Mode by default so a real reminder is never silently batched.

The Pomodoro feature runs focus blocks (default 25 min) followed by breaks (5 min short, 15 min long every 4th cycle). You pick a preset, click Start, and Omniscio drives the run through focus → short-break → focus → … → long-break → … → complete, chiming at each phase transition. Local-only, no AI, no cost.

Pomodoro is rendered as a tab of the Alarms virtual hub — it shares the sound picker, foregrounding helper, and Focus Mode interplay with Alarms, so opening the Alarms project in the sidebar and clicking the Pomodoro tab at the top of Pane 2 is the way in.

Starting a Run

1. Open the Alarms virtual hub in the Omniscio sidebar.
2. Click the Pomodoro tab at the top of Pane 2.
3. Click the Play button on any preset row (or select the preset and click Start in Pane 3). The countdown begins immediately in the Session view.

If the preset has a Default project set, the run is automatically attributed to that project for stats.

Presets — What You Can Customize

Each preset is a named bundle of durations plus optional alarm overrides. The editor in Pane 3 lets you set:

• Name, focus minutes, short-break minutes, long-break minutes.
• Cycles until long break — after this many focus blocks, the next break is long instead of short.
• Total cycles — or "unlimited" to run until you click End.
• Default project — pre-fills the run's project when you Start from this preset.
• Advanced overrides — per-preset sound, assertiveness (modal / banner / silent banner), bring-to-foreground, pierce-focus-mode. Each defaults to "inherit the global default" so a single Settings change propagates everywhere you didn't customize.

Schedules — Auto-Start at a Wall-Clock Time

Click the calendar-clock icon in the Pomodoro tab's header to open the Schedule editor. Each schedule row is a recurring "every Monday at 09:00, start the Deep Work preset" rule. You can also set a project override and a label per schedule.

30 seconds before a scheduled minute, a pre-start toast appears so you can prepare or dismiss. At the scheduled minute, Omniscio atomically claims the slot and starts the run — provided no other run is already active. (Manually-started runs always win — schedules don't displace.)

Schedules are idempotent across restarts: a "last fired minute" marker prevents a re-launched Omniscio from double-firing the same minute. If Omniscio crashed at 8:59:30, the resolver still fires the 9 AM schedule when it restarts at 9:00:05.

The Master Kill Switch

Settings → Notifications → Alarms → Pomodoro: Enable schedules (on by default) is the toggle to silence the entire scheduler without deleting your schedule rows. Vacation weeks, deadline crunches, "just leave me alone for a few days" — flip it off and no pre-start toasts, no auto-starts. Manual runs continue normally.

Each schedule row also has its own Enabled toggle so you can mute just one without flipping the master.

Foregrounding (an important deliberate split)

Alarms default to bring Omniscio to the foreground when they fire (an alarm exists to interrupt). Pomodoro phase transitions deliberately default to do NOT bring to foreground — a 5-minute focus block ending shouldn't yank you out of your deep-work window. The chime and tray badge are enough.

You can override per-preset in the Advanced section, or flip the global default at Settings → Notifications → Alarms → Pomodoro: Phase changes bring Omniscio to foreground.

Auto-enable Focus Mode during focus blocks

Pomodoro can drive Focus Mode automatically so notifications batch up while you're in a focus block and release during breaks. When you click Start, Omniscio turns Focus Mode on; at the focus → break transition it turns Focus Mode off (delivering any queued notifications); at the break → focus transition it turns Focus Mode back on; and when the run ends it restores Focus Mode to whatever state it had before the run began.

Manual-toggle protection. If you click the bell icon yourself during a Pomodoro run — in either direction — Omniscio stops managing Focus Mode for the rest of the run and leaves your choice intact. The next time you Start a run, automatic management resumes from scratch.

Three places to control it:

• Global default at Settings → Notifications → Alarms → Auto-Enable Focus Mode During Focus Blocks. Off by default — opt in here once and every preset inherits it.
• Per-preset override in the preset editor's Advanced section. Three options: "Use default" (inherit the global setting), "Yes" (force on for this preset), "No" (force off for this preset). Useful when most presets should silence notifications but your "Light Reading" preset shouldn't.
• Mid-run manual toggle — click the bell icon and the rest of this run uses your manual choice, no Pomodoro management.

If Omniscio crashes mid-run, the next boot restores Focus Mode to the pre-run state (the boot-time crash sweep), so a crashed Pomodoro can't leave you permanently in Focus Mode.

Stats — Today, This Week, by Project

Click the history icon in the Pomodoro tab's header to open the History view. Three rollups:

• Today — total focus seconds completed today (local timezone, resets at local midnight).
• This Week — total focus seconds since the most recent Monday 00:00 local.
• By Project — rows for each project that has any focus seconds this week, sorted by weekly total. Runs without a project aggregate under "Unassigned".

Above the rollups, the Today's planned vs actual block shows your day's planned focus minutes (sum of today's enabled schedules × their preset's focus minutes) alongside actual focus seconds completed — a quick "did I stick to the plan?" glance.

Pause and Resume

The Session view has Pause and Resume buttons. Pausing freezes the countdown; resuming extends the phase end time by the pause duration so the remaining time is exactly what you saw when you paused.

Caveat: pause state is in-memory. If you close Omniscio while paused, the run is marked as crashed on the next boot. Use the End button before closing if you want a clean record.

The Gmail integration lets your AI agents access your email. This enables workflows like "check my emails and summarize what needs attention" or "draft a reply to this email thread."

Setting Up

1. Go to Settings → Email & Summaries → Gmail.
2. Click "Connect Gmail."
3. Sign in with your Google account and grant permissions.
4. Omniscio can now read your Gmail when you ask it to.

What You Can Do

• Read your inbox right in the sidebar — opening Gmail drops you on the message you were last reading (or the first one in your inbox) so you start on an email instead of an empty pane
• See every file attached to an email right on its message — each shows its name and size, and one click saves it to your computer (on a phone it downloads or opens your share sheet; a file over 5 MB is best saved from your computer). Omniscio never opens an attachment for you, and pictures already shown inside the email, like a signature logo, aren't listed twice
• Filter your inbox by category using the dropdown above your message list: All, Sent, Primary, Updates, Social, Promos, or Forums; Sent lets you confirm an email actually went out. Sending, replying, or forwarding now shows a quick confirmation toast
• Ask agents to check your recent emails
• Get email summaries
• Have agents draft replies
• Use emoji input in any compose box — type : then two characters for a quick autocomplete, or click the smiley button for a full picker with search and recents
• Use Email Cleanup — an AI-guided walkthrough that scans your inbox, sorts senders into groups (promotions, newsletters, receipts, people…) with a recommended action for each, and clears the clutter one tap at a time (every action confirmed first). Ask it in plain words too — “unsubscribe from all the promos but keep anything from my bank” — and it proposes the batch for you to approve. Reach inbox zero fast; also reachable from the ✨ button inside the Gmail inbox
• Set up the Email Summarizer for automatic email digests

Omniscio also protects anything you're writing: while you're composing, replying, or forwarding, your draft saves automatically as you type — first to your device for instant crash protection, then to your Gmail Drafts folder so it follows you to another device. If you close the compose window or Omniscio itself before sending, a banner in the Gmail inbox lets you pick right back up (or discard the draft) next time you're there, and the compose window shows a quiet “Saving…” / “Saved” note so you always know your work is safe. Each connected account keeps its own draft, and nothing is cleared until you send or choose to discard it.

Connect Google Calendar so your AI agents can see your upcoming events. This enables smart scheduling and time-aware responses.

Setting Up

1. Go to Settings → Google Workspace and find the Calendar card.
2. Turn on Calendar Integration, then click "Authenticate."
3. Sign in to Google and grant access.

What You Can Do

• See daily agendas in the Calendar virtual hub
• Ask agents about your schedule
• Get automated morning briefings with your day's events

The Calendar View

Click Calendar in the sidebar to open a full calendar panel. Tabs at the top switch between three views: Agenda (a scrolling list of what's coming up, the default), Week, and Month. The panel covers the next 90 days and the header shows an event count plus New Event and Refresh buttons. Click any event to see its details in a small popover, with Edit and Delete buttons (Delete asks for confirmation first). In the Week view you can also drag an event to a new time, or drag its bottom edge to make it longer or shorter.

Creating and Editing Events

New Event (or Edit on an existing one) opens a full event editor, not just a title box:

• Title, start and end times, and an All day switch
• Location and description
• Attendees: invite people by email address
• Repeat: daily, weekly, monthly, yearly, or a custom rule (every N days/weeks/months, specific weekdays, an end date)
• Reminders: keep your calendar's defaults, or add custom ones (notification or email, from 5 minutes to 2 days before)
• Which of your calendars the event is saved to

Calendar AI Chat

The bottom of the calendar panel has a chat bar. Ask it things like "what does my Thursday look like?" or tell it "move my 3pm with Sam to 4pm", and the AI reads and edits your real calendar to answer. Every action it takes shows as a small labeled badge on the reply (Created, Updated, Deleted), so you can see exactly what it did. Press the arrow to expand the conversation, Esc to collapse it. The chat needs an API key account in Settings; it tells you if one is missing.

Today and Tomorrow in the Sidebar

While the Calendar project is selected, the sidebar shows a compact Today and Tomorrow list: start time, title, and location for each event. The event happening right now (or starting within the next two hours) is highlighted, past events are dimmed, and a location that is a link (like a video call URL) is clickable.

PM Due Dates in Calendar

If you have Mission Control enabled, your board items' due dates and sprint timelines can appear as an overlay on the Calendar. They sit alongside your Google Calendar events, color-coded by board, with overdue items highlighted in red. Click a PM event to see the board name and due date, plus an "Open in PM" button that jumps to the Mission Control panel. Toggle it at Settings → Mission Control → "PM due dates in Calendar" (on by default).

If Google Access Stops Working

Google sign-ins can expire or be revoked. If that happens while you're creating an event from Quick Launch, Omniscio explains the problem in plain language and shows a Reconnect Google button: click it, approve the sign-in, and Omniscio retries creating the same event automatically, nothing to retype. The same one-click recovery covers email: if a reply can't send because your Google connection went stale or is missing send permission, Omniscio drops a “can't send email — reconnect Google” card in your inbox with a Reconnect Gmail button, and clears it the moment sending works again — your draft stays safe in the meantime. Supermail tells you too: when it reads your mail through your Omniscio Google connection and that sign-in has expired, its sign-in screen says so and shows a Reconnect Google button — approve the sign-in and your inbox comes back. On your phone it asks you to reconnect Google in Omniscio on your computer instead, because approving the sign-in opens a browser there. Anywhere else, you can reconnect from Settings → Google Workspace, or from the Reconnect button on the Google Account card in Settings, by authenticating again.

Once the plugin is installed, the GitHub integration enriches your projects with pull request and issue data from your repositories.

You can turn the legacy built-in version back on temporarily in Settings → Lab → Built-in GitHub (legacy), but the plugin is the supported path.

Setting Up

Do this after installing the GitHub Integration plugin from the Marketplace (see the note above).

1. Install GitHub CLI and run gh auth login once in a terminal to sign in.
2. In Omniscio, turn the built-in GitHub integration on at Settings → Lab → Built-in GitHub (legacy), then switch its notifications on there. (The supported path is the GitHub Integration marketplace plugin — see the note at the top of this page.)
3. If the status reads "Authenticated but missing notifications scope", click Grant Access. A modal pops up with a one-time XXXX-XXXX code and a Copy button.
4. Your browser auto-opens to https://github.com/login/device. Paste the code there and authorize the notifications scope.
5. The modal closes when GitHub confirms. The status flips to "Connected as <your-username>" and PR / issue / CI notifications start showing in the sidebar.

Click Cancel (or close the modal) to abort mid-flow — Omniscio kills the gh child process and silently re-syncs status, no error toast.

Features

• PR tracking — See the open pull requests that involve you
• PR detail panel — View PR title, description, checks status, review comments, and changed files
• Issue tracking — See issues you're involved in (assigned, mentioned, or subscribed)
• GitHub virtual hub — Aggregated view of all GitHub activity across hubs

To actually work through pull requests and issues (review, approve, merge, comment, close), install the GitHub Integration marketplace plugin from Settings → Plugins → Browse Marketplace.

16

The Pull Requests tab in the left sidebar gathers the GitHub pull requests that involve you, from every repository your GitHub account can access, into one place. You can read a pull request, look through its changed code, follow its checks, hold line-by-line review conversations, approve it or ask for changes, and merge it, all without opening github.com.

Getting Started

1. Click Pull Requests in the left sidebar.
2. If GitHub isn't connected yet, the tab shows a Connect GitHub button that walks you through the one-time GitHub CLI install and sign-in. Already signed in? Your pull requests appear right away.

The List

Filter chips at the top switch between All (every open pull request that involves you, the default), Created (ones you opened), and Review (ones waiting on your review). When the list spans several repositories, a dropdown narrows it to a single repo. Rows are sorted newest first.

Reading a Pull Request

• Description and activity: the write-up, the comments, and past reviews.
• Changed files: every touched file with its added and removed line counts. Expand a file to read its changes line by line.
• Line comments: review conversations appear directly under the line they were left on (older ones whose line has since changed are grouped under an Outdated heading, so nothing is lost). Hover a line and click the + to start a new conversation, reply inside any thread, and mark it resolved once it's handled.
• Checks: under the overall pass/fail badge, a per-check list shows each CI check with its own state and a link to its logs. A finished check that didn't pass has a Re-run button, so you can retry it without leaving Omniscio.
• Browse files: a folder tree of the entire repository as that pull request sees it. Click any file to read its full contents, with word wrap, copy, and an open-on-GitHub link.

Acting on It

The action bar lets you Comment, Approve, Request changes, or Merge. Merging asks you to confirm first, because it is the one step you can't take back; the others apply right away and confirm with a small message.

Pull Requests Waiting on You, in Your Inbox

Optionally, pull requests that are waiting on you can show up as rows in your Inbox: ones where your review was requested, and your own ones that got changes requested. Turn it on in Settings → Lab → PRs assigned to me in Inbox. Omniscio then checks GitHub quietly in the background; a row clears itself the moment that pull request stops waiting on you.

The built-in GitHub Issues tab and its New issue button — the in-core surface for browsing, commenting on, and creating GitHub issues — have been removed.

The PR Merge Queue pulls together every open GitHub pull request across the repositories you connect, sorts each one into a lane that tells you how safe, risky, or batchable it is, and lets you hand the merge prep to an AI agent with a single click. The agent works in an isolated copy of your project: it rebases the branch, resolves conflicts by understanding what each side intended, and runs your tests — then stops with that copy ready for you to merge, or tells you why if it can't get it ready safely. It never touches your main branch and never pushes anything — you do the final merge yourself, with the hard work already done.

Setting Up

1. Enable it under Settings → Features → PR Merge Queue.
2. Open Settings → PR Merge Queue and click Add a repo. Pick one of your projects — Omniscio reads its local Git settings and fills in the GitHub repository, the base branch, and a name for you.
3. Open the queue from the toolbar button or the sidebar. Each repo's pull requests appear in lanes such as Fast, Standard, Batch, Risky, and Escalate.
4. Click Merge on any pull request to spawn an agent that prepares the merge in an isolated copy of your project — rebasing, resolving conflicts, and running tests — then stops, ready for you to do the final merge. The merge session spawns in the background — it appears in your sidebar and a toast gives you an Open button to watch it, but it won't pull you away from whatever you're working on. When it finishes, its message to you opens with a plain-English summary of what the pull request does — a few bullet points, plus anything to watch out for — so you can decide whether to merge without reading any code.

What's New

It explains each pull request in plain language

When a merge agent finishes, its final message now opens with a plain-English summary of what the pull request does — a few bullet points in everyday words, no code or jargon — followed by a short "anything to watch out for" line, with the technical details below. So you can read what a change actually does and decide whether to merge it, without digging through the code. (This is the merge agent's own write-up — separate from the one-line summary shown on each pull-request card.) The same applies when several are merged together (a summary per pull request), and even when a merge can't be finished (it explains what it was trying to do and what stopped it).

It refreshes in the background

You no longer have to click Refresh to see the latest pull requests. Once the feature is on and you have at least one repo set up, Omniscio quietly checks your repositories on a timer (every 10 minutes by default, adjustable in settings) and remembers what it found. So when you open the queue it appears instantly with an "Updated 3m ago" note, and it keeps updating itself while you have it open. The manual Refresh button is still there whenever you want to pull the very latest right now.

Your Inbox tells you which pull requests need you

When a pull request needs a person to look at it — or its merge got stuck — your Inbox shows one line for that pull request: an amber "PR #42 needs review", or a red "PR #42 merge stuck". Dismiss a line to archive that pull request (middle-click it on a computer, or tap the dismiss action on your phone) — it disappears from the queue just like archiving an email, and you can always bring it back with Show archived → Unarchive. Click a line instead to see that pull request's queue right there in your Inbox, without leaving it. Snoozing works per repo (it tucks away all of that repo's lines until the snooze ends), and a line clears itself the moment that pull request is no longer waiting — for example, the instant you start merging it, its “needs review” line drops off right away, no need to wait for the merge to finish.

Add your own merge instructions

You can give the agent extra guidance two ways:

• Standing instructions for a repo — a box in that repo's settings ("Standing merge instructions"). Whatever you write there is included with every merge for that repo. For example: "Always squash-merge."
• Instructions for a single merge — each pull request (and the batch button) has an optional Add instructions link that opens a small box for notes that apply to just that one merge.

Your guidance is treated as advice that helps the agent — it never overrides the built-in safety rules that keep a merge from doing something dangerous.

Edit the merge prompt

Those instructions are advice layered on top of the agent's built-in rules. If you'd rather change the rules themselves, you can edit the whole prompt the merge agent receives. In Settings → PR Merge Queue → Merge prompts there are two editable boxes — one for a normal single pull-request merge, one for a batch. Each starts out showing the built-in default; edit it and it saves when you click away, and a Reset to default button puts the safe built-in version back. Details about the pull request (its number, branch, and so on) drop in wherever you leave a {{placeholder}} — the list under each box shows which ones you can use. Each repo can also have its own override under Override the merge prompt for this repo (advanced) in that repo's settings.

Merge a group of related pull requests together

When Omniscio spots several pull requests that don't touch the same files, it groups them in the Batch lane and shows a Merge batch together button (it appears when there are two or more). Click it and a single agent merges the whole group in the right order, checks that each one landed cleanly, and runs your tests once at the end — instead of running a separate merge (and a separate test run) for every pull request. You can add notes for the whole batch with the same Add instructions link.

Dismiss pull requests you don't care about — with the keyboard, too

Some open pull requests just aren't yours to deal with — a long-lived draft, someone else's work-in-progress, something you've decided to leave for later. Each card has an Archive button that hides it from the queue (without closing it on GitHub) and stops it counting toward the "needs review" nudge in your Inbox; a Show archived toggle brings them back, each with an Unarchive button. You can now do all of this from the keyboard: the queue highlights a card when you open it, the arrow keys move the highlight between cards, and your archive shortcut (E or Ctrl + W by default) archives the highlighted one and jumps to the next — so you can clear several in a row without touching the mouse.

Every night, the PR Janitor looks over your open GitHub pull requests and closes the ones whose changes have already fully reached GitHub's copy of your main branch — leaving a short comment explaining why. Anything it isn't sure about, it leaves open rather than guess. If a pull request is stuck on a merge conflict, it starts a session to fix that too, on its own. It never touches a pull request unless that pull request's own changes are already safely on GitHub.

What the numbers mean

After each run, six counts sum up what happened:

• Closed — pull requests closed tonight because their work was already on GitHub.
• Queued to close — already finished, but so far only on Omniscio's own copy of the branch. These close themselves the moment that work actually reaches GitHub — nothing is ever closed early just because Omniscio thinks it's done.
• Stuck — blocked by a merge conflict. The Janitor kicks off a fix for these by itself.
• Needs you — the one count that's actually asking for your attention: something the automatic fix couldn't resolve, or a decision only you can make.
• Could not determine — the Janitor couldn't tell for sure what happened, so it left the pull request open instead of guessing.
• Skipped — left alone on purpose, usually because someone is actively working on it.

Click any past run in the list to see exactly which pull requests it looked at that night and what it decided about each one.

Dry run and Pause

• Dry run — ask the Janitor what it would do right now, without closing anything. Handy for checking in between its nightly runs. A full pass can take a while (up to about an hour on a large backlog), so this button doesn't sit and wait for it — it lets you know the report is on its way rather than leaving you staring at a spinner.
• Pause / Resume — turn tonight's (and every future) run off, or back on. While paused, nothing runs until you resume it.

When a coding-agent session finishes a change that affects the UI, PR Visual Evidence captures a before/after screenshot of the changed screen and attaches it where the change is reviewed — your private Omniscio session by default, and the public GitHub pull request only when the repo explicitly opts into that a second time. It's the automated version of what a careful engineer does by hand: show the screen before and after, side by side, so a reviewer sees the visual diff at a glance.

How it works

1. An agent finishes a UI-affecting change on an opted-in repo and runs the bundled capture step as its last action.
2. Omniscio rebuilds the pre-change version of your app in a throwaway copy, launches it with your preview command, and screenshots the changed screen — that's the real "before".
3. It screenshots the same screen on the new version (the "after"), confirms the two actually differ (otherwise it skips, so you never get a misleading identical pair), and hands both to Omniscio.
4. Omniscio attaches the before/after to your private session. If — and only if — the repo turned on public posting and the branch has an open pull request, it also posts them as one comment on that PR (as short-lived links that expire in about four hours), and never posts a duplicate if you run it again.

What Omniscio enforces for you

• Private always, public only on the second opt-in. The before/after lands in your session first; it reaches a public PR only when you deliberately turned that on.
• It's your repo's PR, or nothing. The pull request is auto-detected from your own branch on your own repository and verified server-side — a session can never post to some other project's PR.
• Test data only. Capture runs against a running app, so it uses seed/test data and skips when it can't be sure — an "after" shot could otherwise reach a public PR.
• Never in the way. The whole thing is best-effort: a missing preview, a build hiccup, or no PR just means the visual is skipped and logged — it never blocks, delays, or fails your actual code change.

AgentMail provides your AI agents with their own email addresses that can receive mail. This enables scenarios where external systems or people can email your AI agent directly, and the agent can process and respond to those emails.

Setting Up

1. Go to Settings → Email & Summaries → Inbound.
2. Enter your AgentMail API key.
3. Create one or more agent email inboxes.

Use Cases

• Receive automated reports from CI/CD systems
• Let clients email questions that your AI agent processes
• Forward specific emails for AI analysis
• Use with the Email Summarizer for automated processing

The SMS integration lets you send and receive text messages to interact with your AI agents from your phone. This is useful when you're away from your computer but need to respond to an agent's question.

Choosing a provider

SMS runs through your own Android phone — there is no Twilio account and no separate SMS number to rent. Pick how the texts travel under Settings → Channels → SMS:

• Pushbullet (the default) — texts are mirrored through the Pushbullet cloud using the Pushbullet app on your phone. Handles pictures and GIFs. Free Pushbullet accounts cap SMS at 100 messages a month.
• Native (my phone) — texts go straight over your phone's real SIM, through a free open-source gateway app you run yourself, so no third-party cloud is in the path and there is no monthly cap. Text-only for now.

You can switch between the two at any time; your conversations and history are shared, so nothing is lost.

Setting Up

Pushbullet (the default)

1. Go to Settings → Channels → SMS and leave the provider on Pushbullet. A three-step wizard appears: Token → Device → Connect.
2. Install the Pushbullet app on your Android phone and sign in. On a computer, open pushbullet.com/account, scroll to Access Tokens, click Create Access Token, and copy the string it gives you (it starts with o.).
3. Paste the token into the wizard and click Save & Verify. Omniscio verifies and encrypts it, then lists your devices — pick the phone marked as SMS capable.
4. Wait for Connected. Text yourself from another number and the message should appear in your Inbox within a few seconds.

Native (my phone, no cloud)

1. On your Android phone, install the free SMS Gateway for Android app (from F-Droid or the project's GitHub releases — it is not on the Play Store). On Android 15 and newer, open its App info → ⋮ → Allow restricted settings once, so it can keep its local server running in the background.
2. Run the app in Local Server mode. It shows an address such as http://192.168.1.5:8080 together with a username and password.
3. In Omniscio, switch SMS provider to Native (my phone) and fill in that address, username and password, plus a webhook secret of your own choosing.
4. Copy the Inbound webhook URL Omniscio shows into the phone app's webhook settings, using the same secret. This needs Mobile Access switched on, so your phone can reach Omniscio.
5. Test it — text yourself from another number, then send a reply from Omniscio.

Off Wi-Fi: the local address only works while your phone and computer share a network. To send and receive while you are away, put both on a private tunnel such as Tailscale, or use the phone app's own cloud relay.

How It Works

• When an agent needs your attention, Omniscio can send you a text message
• Reply to the text to respond to the agent
• SMS conversations appear in your Omniscio inbox
• A conversation leaves your inbox once you deal with it: reply, tap Mark as read, or archive it. Simply opening a thread just marks it read (clears the unread dot) but keeps it in your inbox until you dispose of it, so an unread text is never stuck outside your inbox.
• If SMS is on but stops working (a saved key gets reset, or the connection drops for a sustained stretch), Omniscio raises an “SMS is offline” alert in your inbox plus a desktop notification, and clears it automatically once SMS reconnects. Turn it off under Settings → Channels → SMS → “Alert me when SMS goes offline”.

Email Cleanup scans your Gmail inbox and has AI sort every sender into a category with a recommended action and a reason. The main way to clean up is a chat box — ask in plain English ("unsubscribe from all promos, keep anything from my bank") and it proposes the changes for you to approve. Prefer to do it yourself? Open the collapsed Browse by category section to walk through it one group at a time. Every clean-up is a single confirmed batch — one bulk archive, one toast, one undo — so you can clear hundreds of emails in a few taps without ever losing something you meant to keep.

Opening It

Click Email Cleanup (the ✨ sparkle row) in the left sidebar, or hit the ✨ Clean Up button in the Gmail inbox toolbar — both open the same walkthrough.

What You Can Do

• Chat to clean up — the main surface: ask in plain English ("unsubscribe from all promos, but keep anything from my bank") and it proposes an action batch for you to approve. Tappable example requests get you started, and the chat never executes anything on its own (a chip only fills the box, it never sends).
• Auto-scan on open — it reads your inbox and shows a summary ("N emails from M senders").
• Browse by category — open the collapsed section to work through it by hand: Promotions, Newsletters, Receipts, Social, Updates, People, and Likely-spam, each with the AI's recommendation and the reason behind it.
• Clean up a whole category — one confirmed batch (a single bulk archive with one undo). Senders you mark keep are never touched, and nothing happens without a confirm dialog.
• Drill into individual senders when you want finer control than the whole group.
• Classic layout — prefer a flat per-sender list (Archive / Unsubscribe + Archive / Never again)? Turn on Settings → Email & Summaries → Gmail → Classic Email Cleanup Layout.

The Email Summarizer watches your AgentMail inboxes for incoming emails and automatically creates AI sessions to analyze and summarize them based on rules you define.

Creating Rules

1. Go to Settings → Email & Summaries.
2. Click "New Rule."
3. Select which AgentMail inbox to watch.
4. Define matching criteria (sender, subject keywords, etc.).
5. Write a prompt telling the AI how to handle matching emails.

Use Cases

• Summarize long email threads into action items
• Categorize incoming support emails by urgency
• Extract data from structured email reports
• Draft reply suggestions for common email types

The RSS integration polls a list of feed URLs and surfaces each new entry as its own row in the unified inbox — alongside Gmail, SMS, Slack, Telegram, and webhooks. It's a read-only channel: triage entries like any other inbox row (read, snooze, archive), or let an automation react to them. Each entry shows the title, source feed name, author, and a sanitized preview — click in for the full HTML.

Setting Up Feeds

1. Go to Settings → Channels → RSS and toggle RSS Integration on.
2. Paste a feed URL (for example, a blog's /feed, Hacker News' news.ycombinator.com/rss, or a YouTube channel's /videos.xml). Omniscio validates the URL before adding it — feeds pointing at private/internal addresses are refused for safety.
3. Pick a refresh interval per feed (5–1440 minutes) and a retention window (30 days up to 2 years — older entries are permanently deleted).
4. Rename or delete feeds inline. Use Refresh All to force a poll right now.

Daily Digest

If you subscribe to a lot of feeds, the inbox can get noisy. Enable Daily Digest in the same settings panel and Omniscio will use Claude to produce a single once-a-day briefing that picks out the items most likely to matter to you. Two knobs:

• Delivery Time — when the digest gets generated each day (defaults to 7:00 AM local time).
• Your Interests — free-form keywords or topics (e.g. "AI, startups, privacy, cybersecurity") that the digest uses to prioritise articles.

Each digest lands as its own row in the inbox — a separate source from the raw RSS entries, so you can leave the firehose collapsed and just read the morning summary. Hit Generate Now in settings to produce one on-demand, or expand Recent Digests to see the last 10 with read/unread status.

Pair with Automations

RSS entries flow through the standard Automation engine. Match the title or body against keywords to forward via email, route into the Daily Digest section, or trigger a Claude session to react.

The Telegram integration signs you in to your real Telegram account (not a bot) and surfaces every conversation as a row in the unified Inbox, in real time. Replies you send go out as you — your own name and account — so it works exactly like the official apps, just triaged next to your email and texts.

Telegram is powered by the mtcute engine by default — the maintained MTProto client library (the Telegram via mtcute engine option) — with the legacy GramJS engine kept as a reversible opt-out in Settings → Channels → Telegram engine. Switching engines re-signs you in once.

Telegram Bot is a separate thing, and it is worth knowing which one you want. The integration above brings your own account's conversations into the Inbox so you can read and reply to them. Telegram Bot goes the other way: you create a bot with Telegram's BotFather, pair it once, and then message that bot to talk to your agents from your phone. Each chat — or each topic in a forum group — gets its own session, so several conversations stay separate. Only people you have allow-listed can reach it. The two share nothing but the name: setting one up does not set up the other.

Setting Up

1. Go to Settings → Channels → Telegram and click Sign in.
2. Enter your phone number. Telegram sends a login code to the official Telegram app on another device — type it in.
3. If you have two-factor enabled, enter your cloud password. It's verified on your own machine and never leaves it.

Your session is encrypted and stored locally; signing out clears it.

What You Can Do

• Triage every chat as an inbox row; open one for a full conversation panel with scroll-back history.
• Reply inline — messages send from your own account. Edit, delete, forward, archive, mark-read, mute, and snooze like any other channel.
• See media — photos, voice notes, and documents load on click.
• Automate — run Telegram messages through Automations / auto-replies and the Daily Digest, just like Gmail or SMS.
• Search — Telegram messages show up in global search (Ctrl+K) alongside every other channel.

Telegram Bot — talk to your agents from Telegram

Telegram Bot is a separate channel from the account integration above, with its own toggle in Settings → Channels → Telegram Bot. The account integration signs in as you and brings your chats into the Inbox; the bot channel never touches your account — it runs a bot you create, and it exists so you can chat with your Omniscio agents from your phone.

1. In Telegram, open @BotFather, send /newbot, and copy the token it gives you.
2. In Omniscio go to Settings → Channels → Telegram Bot, turn the channel on, and paste the token. Omniscio verifies it with Telegram and shows “Connected as @yourbot”.
3. Message your bot. The first time, it replies with a pairing code and the request appears in that same settings card (and as an inbox alert) — click Approve.

• One chat, one session. A direct message, a group, or a forum topic inside a group each map to their own agent session. Follow-ups continue that session; send /new to start a fresh one.
• Replies land back in the same place. When the agent finishes its turn, the answer is posted into that chat (or the exact forum topic), split across messages when it's long. While the agent works, the chat shows “typing…”.
• Survives restarts. The chat-to-session links are stored on disk, so quitting and reopening Omniscio picks up where it left off without replaying old messages.
• Uses your normal agent setup. Sessions spawn through the same path as any other session, so whichever Claude sign-in you already use applies. The bot never reads or copies a Claude credential.

The Webhook integration runs a small local HTTP server that accepts incoming POST requests and drops each one into the unified Inbox — alongside Gmail, SMS, Slack, Telegram, and RSS. Point any service that supports outgoing webhooks at it (GitHub, Stripe, Zapier/Make, CI/CD, uptime monitors, your own scripts) and their events show up as messages you can read, route, or react to. It's receive-only — there's nothing to reply to.

Setting Up

1. Go to Settings → Channels → Webhooks, turn on Webhook Integration, and pick the port the local server listens on.
2. Click Add a webhook source. Each source gets a unique URL slug and an auto-generated bearer token.
3. Copy the resulting URL (http://<your-host>:<port>/<source>) into the service that will POST to it, and include the bearer token in an Authorization header.

What You Can Do

• Run multiple named sources, each with its own URL and token, grouped separately in the inbox.
• Read each payload in the inbox: request method, headers, pretty-printed JSON body, timestamp, and the sender's IP.
• Route with Automations — keyword-match the body, forward to email, or fold it into the Daily Digest.
• Trigger a session — have an incoming event kick off a Claude session to react to it.
• Search webhook bodies via global search (Ctrl+K).

The Zapier integration injects your personal Zapier MCP server into every Claude session automatically. Once enabled, agents can trigger Zaps, read data from connected apps, and manage workflows across services like Google Sheets, Slack, Shopify, and thousands more — all without any per-session setup.

Getting Started

1. Go to mcp.zapier.app and sign in (or create a free Zapier account).
2. Create a new MCP server — Zapier generates a unique server URL for you.
3. In Omniscio, open Settings → Channels → Zapier.
4. Enable the toggle and paste the server URL.
5. On Zapier's configuration page, enable the specific app actions you want your sessions to use.

What Sessions Can Do

With Zapier enabled, Claude sessions get access to tools for discovering available app integrations, enabling or disabling specific actions, reading data from connected apps, and triggering write actions or Zaps. You manage which apps and actions are available from Zapier's web UI.

Auth & Security

Zapier uses URL-based authentication — your unique server URL is the credential. Individual app connections (Gmail, Slack, Shopify, etc.) use OAuth managed entirely by Zapier's web UI. Omniscio never touches those credentials.

Rate Limits

Zapier actions count toward your Zapier plan quota (free: 100 tasks/month). There is no separate MCP-specific rate limit.

The Fathom integration runs a local MCP server that wraps the Fathom.ai REST API. Once enabled, every Claude session can search your meetings, read full speaker-labeled transcripts, and view AI-generated summaries — all without leaving the conversation.

Getting Started

1. In your Fathom account, go to User Settings → API Access and generate an API key.
2. In Omniscio, open Settings → Channels → Fathom.
3. Enable the toggle and paste your API key.
4. New sessions now have Fathom tools available automatically.

What Sessions Can Do

• List meetings — browse your recent meetings (title, date, duration).
• Search meetings — find meetings by keyword.
• Read transcripts — full speaker-labeled transcript of any meeting.
• View summaries — AI-generated summary from Fathom.
• List teams — see available teams in your account.

Auth & Security

Your API key is encrypted at rest and never sent to the renderer. It is passed to the MCP server process as an environment variable. Fathom is automatically disabled for Search, Ask Omniscio, and Vault sessions to protect your credentials.

Rate Limits

Fathom allows 60 API requests per minute. If you hit the limit, the agent receives a descriptive error and can retry after a brief wait.

The Microsoft Teams integration connects to Microsoft Graph API via OAuth 2.0 to pull your calendar events that have online meetings attached. Once connected, Omniscio syncs meetings on a configurable interval so your schedule is always up to date.

Getting Started

1. Open the Microsoft Teams panel in the sidebar.
2. Click Connect Microsoft Account. A browser window opens for Microsoft sign-in.
3. Sign in with your work or personal Microsoft account and grant the requested permissions.
4. Your meetings sync automatically. Click Sync Now to trigger an immediate refresh.

What You See

• Meeting list — upcoming and recent meetings grouped by date.
• Meeting detail — organizer, time, attendees, body preview, and a direct join link.
• Settings tab — sync interval, sync window, and disconnect controls.

Auth & Security

Your Microsoft credentials are encrypted at rest and never sent to the renderer. The integration uses OAuth 2.0 with PKCE (no client secret stored). Only calendar read permissions are requested.

Sync Behavior

Background sync runs every 15 minutes by default (configurable from 5 min to 24 hours). The sync window is symmetric — one number sets how far back and how far ahead it looks. It defaults to 14 days each way, and you can raise it to 90 days each way (configurable). Delta queries are used for incremental updates after the initial full sync.

Team Chat is Omniscio's built-in messaging system for your team. Unlike external integrations, it runs inside Omniscio with no third-party dependency. Messages, channels, and presence are synced in real time across desktop and mobile.

Getting Started

1. Open the Team Chat panel from the sidebar.
2. Create or join a workspace. Organization members are added automatically; self-serve workspaces use invite codes.
3. A #general channel is created automatically in every new workspace.
4. Start messaging. Desktop notifications and mobile push are available out of the box.

Features

• Channels — public and private channels for topic-based conversations.
• Direct messages — private one-on-one conversations with presence indicators.
• Threads — reply to any message in a thread to keep conversations organized.
• Reactions — react to messages with emoji.
• File attachments — share images and files in any channel (desktop).
• Search — find messages by keyword across all your channels (desktop).
• Mentions — @mention teammates to notify them directly.
• Presence — see who is online, away, or offline in real time.
• Saved items — bookmark important messages for later reference.
• Webhooks — send automated messages into channels from CI/CD, monitoring, or other tools.

Mobile Access

Team Chat has a Progressive Web App (PWA) for mobile access. Open the Team Chat URL on your phone and add it to your home screen for an app-like experience with push notifications. On iOS, you must install the PWA to receive push notifications.

Workspaces

• Organization workspaces — linked to your Omniscio organization. All org members have access automatically.
• Self-serve workspaces — standalone workspaces created by any user. Share an invite code or link to add members.
• Cross-org connections — connect with users outside your organization for direct messaging.

Privacy & Data

• Messages are stored in Firestore with encryption in transit and at rest.
• You can export your data or delete your account at any time from the Team Chat sidebar.
• GDPR right-to-erasure is fully supported — account deletion removes all your messages, search index entries, presence data, and device tokens.

The AHK integration connects Omniscio with AutoHotkey, a popular Windows automation tool. This lets you trigger Omniscio actions from global hotkeys, even when Omniscio isn't the focused window.

Setting Up

1. Install AutoHotkey v2 on your system.
2. Install the AutoHotkey plugin — open the Marketplace from the sidebar (the store icon), or go to Settings → Plugins → Browse Marketplace. The in-app AutoHotkey editor and Manager were retired, so the plugin is the supported path.
3. Configure your AHK scripts to communicate with Omniscio via the CLI API.

Use Cases

• Global hotkey to start a new AI session from anywhere
• Keyboard shortcut to send selected text to an AI agent
• Macro to trigger a recipe run

The Canva integration gives your Claude sessions direct access to Canva's design platform. Once enabled, agents can create designs from prompts, search your design library, export to common formats, and manage assets and folders — all without leaving the conversation.

Getting Started

1. In Omniscio, open Settings → Channels → Canva.
2. Enable the toggle.
3. The first time a session uses a Canva tool, your browser will open for OAuth sign-in — no API keys or manual setup needed.
4. Once authenticated, all new sessions automatically have Canva tools available.

What Sessions Can Do

• Create designs — generate new designs from text prompts using Canva's AI-powered creation.
• Search designs — find designs in your Canva library by keyword.
• Export designs — download designs as PDF, PNG, or JPG.
• Manage assets & folders — organize your design files and assets.
• Use brand templates — create from your organization's templates (requires Canva Pro or Enterprise).

Auth & Security

Canva uses OAuth authentication — when a session first tries to use a Canva tool, your browser opens automatically for sign-in. The token is managed by the Canva MCP server and refreshed automatically. Omniscio never stores your Canva password.

Limitations

• Paid features — brand templates and some advanced creation features require a Canva Pro or Enterprise plan.
• Rate limits — Canva enforces rate limits: approximately 20 requests/minute for design creation and 100 requests/minute for read operations (search, list, export).
• Individual accounts only — the integration currently supports individual Canva accounts; team/organization-wide setup is not yet available.

What you get

Once enabled, a Ollert row appears in the Omniscio sidebar. Clicking it opens Ollert inside Omniscio with its own navigation sidebar on the left — jump between Boards, Templates, Home, and your workspaces — and the board you pick fills the panel beside it, just like Omniscio's other integrations. Your boards, lists, and cards are all there, and you can work on them without switching to a browser tab. There is no separate Ollert sign-in — Ollert uses your Omniscio account. Just sign in to Omniscio and Ollert is ready; if you are signed out, Ollert shows a “Sign in to Omniscio” prompt instead.

Ollert talks to a separate Ollert backend (not run by Omniscio). Because it uses your Omniscio sign-in, you don't have to log in separately or again after a restart.

Sharing boards and cards

You can copy a link to any board or card to share with a teammate who also has Omniscio:

• Copy link to card — open the card's … menu and choose Copy link to card.
• Copy link to board — use the Copy link to board action in the board header.

Both actions copy a deep link to your clipboard. When your teammate opens that link on a machine with Omniscio installed, Omniscio opens straight to that card or board — no searching, no copy-pasting board names.

How to turn it on

Ollert is gated on ollertEnabled (off by default; set for existing installs by the data-presence migration). There is no Settings toggle — the old “Enable Ollert” control was removed, and Settings no longer lists Ollert. While off, nothing Ollert-related loads and the sidebar row is hidden.

What you get

Once enabled, a Trello row appears in the Omniscio sidebar. Clicking it opens a kanban board view of your real Trello boards. The left sidebar lists your open boards; select one to see its lists as columns with cards sorted by position. Click any card to open a detail drawer showing the description, checklists, comments, labels, members, and due dates.

Setup

1. Open Settings → Lab and enable Trello.
2. Enter your API Key and API Token from trello.com (avatar → Power-Ups → API Key, then generate a token).
3. Click Test Connection to verify.

Your credentials are encrypted on your device and never leave it.

Working with cards

• Create, update, and archive cards.
• Move cards between lists.
• Add comments and manage checklists.
• Search cards across your boards.

Inbox

Enable Trello Inbox in settings to see cards assigned to you in the unified inbox. Kill switch: AMC_DISABLE_TRELLO_INBOX_POLL=1.

Workflow triggers

Create an automation with the Trello card trigger to fire when cards are created, updated, or move between lists. Kill switch: AMC_DISABLE_TRELLO_WORKFLOW_TRIGGER=1.

CLI routes (for AI agents)

• POST /trello/move-card — { cardId, listId } moves a card.
• POST /trello/create-card — { listId, name, ... } creates a card.

Note

This connects to your real trello.com account. It is separate from Ollert, which is a standalone kanban tool with its own backend.

Connecting

Asana connects with a Personal Access Token: create one at app.asana.com → My Settings → Apps → Developer apps → Personal access tokens. Enter the token and your workspace GID in Settings. Omniscio talks directly to Asana's REST API; nothing goes through a third-party server.

What you get

• Board — your project's sections with tasks, move-to-section, and reorder.
• Inbox — optionally surface "tasks assigned to me" updates in the Omniscio inbox (poll-based, background refresh).
• Workflows — Asana task events (created, completed, updated) can trigger Omniscio workflow automations.
• CLI — move tasks between sections and reorder from the CLI server.

Automation

Asana tasks can trigger Omniscio Workflows on three events: task created, task completed, and task updated. The trigger polls every 2 minutes and fires only on genuinely new changes.

Connecting

Jira connects with your Atlassian account: enter your site URL (like yourteam.atlassian.net), your Atlassian email, and an API token (created at id.atlassian.com → Security → API tokens) in Settings. Omniscio talks directly to Jira Cloud's REST API — nothing goes through a third-party server.

What you get

• Board — your project's columns with drag-to-transition, quick filters, and epic swimlanes.
• Issue detail — edit fields, transition, comment, log work, watch/vote, link issues, and attach files.
• Planning — backlog + sprint management with estimates.
• Summary — a project landing page with status, priority, and assignee breakdowns plus recent activity.
• Docs — a lightweight page tree for project notes, stored locally in Omniscio (Jira Cloud offers no docs write API).
• Inbox — optionally surface "issues assigned to me" updates in the Omniscio inbox.

Agent access (CLI)

AI agents can work Jira issues headlessly through Omniscio's CLI control server: search issues via JQL, view issue details, create and edit issues, transition statuses, and add comments. Mutations are rate-limited. All routes use the same credentials you configured above.

Default project

Set a default project in your Jira settings to pre-select a project when creating issues.

Automation

Jira issues can trigger Omniscio Workflows, and workflow actions can create, transition, comment on, assign, and search Jira issues — so an issue update can kick off a real agent task.

What you get

• Projects & organizations — create projects in your organization, invite teammates by email (even across organizations), and control access with three simple roles.
• Board, backlog & sprints — drag-and-drop kanban with swimlanes, JQL-scoped saved boards, sprint planning, and bulk editing.
• Issues — types, priorities, story points, labels, components, versions, worklogs, watchers, attachments, links, and @mention comments with notifications.
• Automation — project or organization-wide rules ("when an issue is created, if priority is high, assign and comment").
• Reports & docs — a summary dashboard, velocity and flow reports, and a lightweight per-project docs tree.

Agents work your issues

The headline feature: every issue has a Start agent button. Omniscio spawns a real agent session in the project's working folder (you pick the folder the first time), briefs it with the issue's summary, description, and acceptance criteria, and mirrors the session's live status back onto the issue card — Running, Needs You, or Finished — with system comments on the meaningful transitions. You can also bulk-start agents for a whole sprint.

What you get

• Boards and items — create boards with columns, groups, and items. Kanban and table views with drag-and-drop.
• Automations — board-level automation rules with triggers, conditions, and actions (including AI-powered actions that spawn agent sessions).
• Board templates — save a board's structure as a reusable template, deploy templates to new boards, and share via a gallery.
• Calendar overlay — PM due dates and sprint timelines appear on your Calendar alongside Google Calendar events.
• Inbox source — PM notifications flow into your unified inbox.
• CLI access — agents can read boards, create items, and update columns via the CLI control server.

PM Notification Intelligence

An additive layer that routes board changes to your unified inbox as smart notifications — batched per session so a burst of edits becomes one digest instead of a pile of pings, and tuned per board so you hear about what matters and mute what doesn't.

• Session-scoped batching — changes made during one work session are grouped into a single notification rather than one alert per edit.
• Per-board preferences — set each board to normal, muted, or promoted, choose email delivery (off, instant, or digest), and pick a daily or weekly digest cadence.
• Due-date alerts — a background check scans your boards for overdue and soon-due items (within 24 hours) and raises an inbox alert, so deadlines surface on their own.

Session linking

Sessions can be linked to PM items. A linked session shows a PM item chip in the session header with the item's name. Click the chip to navigate to the board; right-click to unlink.

• Start a session from a PM item — the CLI route POST /pm/items/:id/start-session spawns a new session pre-linked to the item, with PM context (item name, board, column values) automatically injected into the agent's briefing.
• Link after creation — any session can be linked or unlinked to a PM item at any time.
• Live status sync — when a linked session's status changes (running, needs you, finished, error), the PM side is notified in real time so the board can reflect agent state.

Cross-entity linking

PM items can also be linked to KMS vault notes, mind maps, flowcharts, and whiteboards. Links are bidirectional:

• From the PM item card — open an item and switch to the Links tab to see all linked entities, grouped by type. Hover over a link to reveal an unlink button.
• From a visual tool — each tool (KMS file tree, mind map toolbar, flowchart toolbar, whiteboard toolbar) has a Link to Board Item action. Click it to search across all boards and items, then select one to create the link.
• Backlink chips — linked items appear as small chips in the header of each visual tool. Click a chip to navigate to the Mission Control board.

AI Intelligence

Three layers work together to give you AI-powered insight across all your boards:

• PM Cross-Board Intelligence — aggregates data across boards into materialized snapshots, tracking lead times, detecting patterns, applying seasonal urgency, and answering natural-language queries. Runs in three computation tiers (hot, warm, cold) with zero LLM cost.
• PM Operations Engine — per-board policies (Auto/Propose/Never) that govern session shepherding, work queue management, shift handoffs, and dashboard health. All algorithmic, zero inference cost. Trusted actions can auto-promote after consecutive approvals.
• PM Intelligence Advisor — the advisory layer that turns raw data into actionable insight: strategic coaching, Monte Carlo completion predictions, policy recommendations, enriched shift handoffs, and a quick-question box for plain-English queries about your projects. Open it from the PM Advisor entry in the sidebar.

How it works

The import framework is a 5-stage pipeline that any source adapter plugs into:

1. Extract — the adapter reads the source (a file, an API, etc.) and produces a canonical intermediate representation (IR) containing boards, groups, columns, items, and column values.
2. Map — columns are mapped to PM column types. Known sources (Monday.com, Jira) have explicit type maps; unknown sources fall back to automatic inference from header names and values.
3. Validate — the manifest is checked for non-empty board names, no duplicate source IDs, valid references, and a 5,000 item-per-board limit.
4. Execute — the entire import runs inside a single SQLite transaction for atomicity. Entity mappings are recorded for re-import dedup.
5. Report — counts of boards, groups, columns, and items created, skipped, or failed, plus warnings and errors.

Conflict strategies

When re-importing data that overlaps with previously imported entities:

• Skip (default) — if the entity was already imported, skip it entirely.
• Overwrite — reuse the existing local ID and replace the data.

CSV import

The built-in CSV adapter reads standard CSV files (RFC 4180), strips UTF-8 BOM markers, deduplicates header names, detects groups from a "Group" or "Section" column, and infers column types from header names and cell values. File size cap: 50 MB.

What you get

• Boards and items — create boards with columns, groups, and items. Kanban and table views with drag-and-drop.
• Automations — board-level automation rules with triggers, conditions, and actions (including AI-powered actions that spawn agent sessions).
• Board templates — save a board's structure as a reusable template, deploy templates to new boards, and share via a gallery.
• Calendar overlay — PM due dates and sprint timelines appear on your Calendar alongside Google Calendar events.
• Inbox source — PM notifications flow into your unified inbox.
• CLI access — agents can read boards, create items, and update columns via the CLI control server.

Session linking

Sessions can be linked to PM items. A linked session shows a PM item chip in the session header with the item's name. Click the chip to navigate to the board; right-click to unlink.

• Start a session from a PM item — the CLI route POST /pm/items/:id/start-session spawns a new session pre-linked to the item, with PM context (item name, board, column values) automatically injected into the agent's briefing.
• Link after creation — any session can be linked or unlinked to a PM item at any time.
• Live status sync — when a linked session's status changes (running, needs you, finished, error), the PM side is notified in real time so the board can reflect agent state.

ContextDock is a separate web app where you collect markdown notes — research, customer transcripts, product specs — into named bundles (hand-curated collections of docs) and lists (tag-driven views over your docs). The Omniscio integration lets you link any bundle or list into a project so every Claude session in that project automatically sees the contents.

Setting Up

1. Mint an API key on contextdock.web.app (workspace settings → API keys → New key with read scope).
2. In Omniscio, reveal the section first — the ContextDock Settings row is hidden by default (2026-08-14). Turn on ContextDock under Settings → Lab → Built-in Apps; the row then appears at Settings → Tools & Maintenance → ContextDock. Paste the key into the input.
3. Click Validate & Save. Omniscio checks the key against the server and shows a green check + your identity label on success.

The key is encrypted via Windows DPAPI and stored locally. If the network is down during validation, Omniscio offers a Save without validating escape hatch — the draft key is written and you can validate later.

Linking Bundles, Lists & Library Docs

1. Open a hub. Click Add Hub Doc (the + Doc button in the docs section).
2. Pick the ContextDock tab (fourth tab, always present).
3. Switch between the Bundles (📦), Lists (🏷️), and Library sub-tabs. Bundles and Lists each have their own search box; the Library tab searches your whole ContextDock doc library and links one doc at a time at a compression level you choose — it works even when you have no bundles or lists.
4. Bundles: click Add to link the whole bundle in one shot at the author-set compression level.
5. Lists: Add index writes a titles-only table of contents for the list. Browse drills into per-doc selection so you can pick which docs go in and at which compression (Original / Key Points / Summary).
6. Library: search your entire doc library, pick a compression level per result, and click Add to link a single doc — the quickest path when you know the doc you want and don't care which bundle or list it lives in.

Each linked item lands as a contextdock-<id>.md file in <project>/.claude/docs/. Every new session in that project automatically sees the contents — folded into the agent's first message, on any engine (Claude, Codex, Gemini, …). A freshly-linked doc reaches the next new session, not one already running.

Background Refresh (hourly)

The picker reads from a local SQLite cache so opening the ContextDock tab is instant. The cache refreshes itself each time you open the picker (whenever it’s more than a few minutes stale) — that on-open refresh is what keeps it current. A light background refresh task also runs hourly (after a 60-second startup delay) as a backstop, re-syncing the bundles, lists, and tags rosters quietly; per-list-detail is fetched on demand when you open a specific list, not on the timer. Docs use a delta-sync watermark once warm so steady-state refresh cost is tiny.

Toggle Background refresh (hourly) at Settings → Tools & Maintenance → ContextDock to turn off even that hourly backstop. The cache still refreshes whenever you open the picker or click the 🔄 icon. The toggle takes effect immediately — no restart needed.

Pinning Bundles

Pin a bundle to keep it at the top of the Bundles sidebar pane. Two ways to pin:

• Hover-reveal pin icon — hover over any bundle row in the sidebar to reveal a small pin icon. Click it to toggle. Pinned bundles show the icon in accent color; unpinned bundles only show it on hover.
• Edit dialog toggle — open a bundle's edit dialog and flip the Pinned toggle (below Shared). Only available in edit mode, not when creating a new bundle.

Pinned bundles sort before unpinned bundles; within each group, bundles sort alphabetically by name.

Previewing, Refreshing, Unlinking

• Left-click a linked row to preview the snapshot's rendered markdown.
• Long-press a row (~0.4s) and drag to reorder it in the Auto Context list — no grip handle needed, and a quick click still previews.
• Right-click → Refresh to re-fetch a linked snapshot from upstream.
• Change compression (per-doc items only) — click the inline compression chip on the doc's row to cycle Original → Key Points → Summary, or right-click → Change compression to jump straight to a level. Either way the swap is atomic.
• Middle-click a row to unlink instantly with a Ctrl+Z undo toast, or right-click → Unlink for a confirmation-dialog path.

Omniscio has a plugin system that lets you add new capabilities to the app. Plugins can add new views, services, and integrations without modifying the core app.

Beyond the built-in plugins below, the plugin platform itself ships two always-on capabilities: an installed plugin can register its own actions as CLI endpoints that Omniscio's AI and command-line interface (CLI) can invoke (Plugin AI-native CLI), and a plugin that requests permission can add its own buttons to the header toolbar and session menus, and navigate the app to a session, hub, or view (Plugin UI access).

When a plugin's AI action would change something — posting, publishing, editing, or deleting — it asks you first: the request lands as an approval card in your inbox instead of running right away. Reads and small, easily-undone actions run immediately. Right on that approval card you can choose “Always allow this action for this plugin” so that one action stops asking; manage or remove these anytime at Settings → CLI Control → Always-allowed actions.

Managing Plugins

1. Go to Settings → Plugins.
2. See installed plugins and their status.
3. Toggle plugins on or off.
4. Configure plugin-specific settings.

Available Plugins

Omniscio currently ships with two built-in plugins:

• Foundry — AI-powered app builder — from idea to deployed project
• RepoGuard — Repository security scanning and health scoring

Plugin Marketplace

The Plugin Marketplace is Omniscio's built-in catalog for browsing, installing, updating, rating, and uninstalling plugins — no manual file copying, JSON editing, or app restarts required. Once enabled (see below), open it from the sidebar (look for the Store icon); it lives as a virtual hub with its own search box, sort dropdown, and category sidebar.

What you can do

• Browse the grid. Each card shows the plugin icon, name, one-line description, average star rating, download count, version, author, and category. An amber Update chip appears on plugins you have installed when a newer version is available in the registry.
• Filter by category. Seven categories run down the left sidebar: All, Planning, Development, Testing, DevOps, Productivity, Other. The count beneath them tells you how many plugins are in whichever category you picked.
• Search and sort the grid. Both sit in the toolbar directly above the plugins — search on the left, sort on the right. Four sort options: Most Popular (downloads), Highest Rated (average stars), Newest (most recent publish), A–Z. On a phone the categories and the results are separate screens, so tap a category first; search and sort are waiting on the results screen.
• Open a plugin's detail page. Click a plugin's name, icon, description, or star rating to see the full description, changelog, permissions, install button, star-distribution histogram, and the latest 20 reviews (with a Load more button to paginate). The card's Install, Update, and on/off controls still do only their own job. Press Back, Esc, or the Browse breadcrumb to return to the grid.
• Read reviews. A plugin's detail page shows its average score, the star breakdown, and the latest reviews. Leaving your own review needs a marketplace sign-in with GitHub — your Omniscio account alone isn't enough, and the regular release of Omniscio doesn't include that sign-in yet, so the rating box tells you rating isn't available rather than showing stars that could never be sent. Reading reviews is unaffected, and you can't rate a plugin you published yourself.
• Check for updates. Omniscio checks the marketplace for newer versions automatically — on launch, every few hours, and when you return to the window — and flags them with the Update chip. To check on demand, click Check for updates at the bottom of the Marketplace sidebar, under your plugin counts: the answer appears right on that line as up to date, 2 updates, or check failed. If it finds anything, the review list opens straight away showing which plugins have an update, what version each moves to, and any new permissions they ask for — and while updates are pending that 2 updates line stays clickable, so you can reopen the list any time after closing it. There's also Check now under Settings → Plugins, which reports the same thing as a notification. Neither installs anything — applying an update is always your click.
• Update in place. A plugin showing the Update chip can be updated with one click — Omniscio atomically swaps in the newer bundle. Your local plugin data (SQLite tables, key-value storage) is preserved.
• Uninstall. Removes the plugin's files and the virtual sidebar project. Your plugin data is kept in SQLite so reinstalling later restores your work.

For plugin developers

If you're building plugins, enable the Plugin Developer Dashboard in Settings → Plugins. It surfaces as a separate sidebar entry showing the SDK quick-start commands, links to the GitHub repo and full documentation, and your submission history (Pending Review / Published / Rejected, with reviewer notes inline for rejections). See Build Your Own Plugins.

Browsing Plugins

Open the Marketplace from the sidebar (look for the store icon). You'll see a grid of available plugins with:

• Search — filter plugins by name or keyword
• Categories — filter by plugin category (productivity, development, integration, etc.)
• Sort — sort by newest, most popular, or highest rated

Installing a Plugin

1. Click a plugin's name or description to see its details — description, permissions, changelog, and ratings.
2. Click Install.
3. If the plugin requires permissions (storage, network, etc.), a consent dialog appears listing what the plugin needs. Review and click Install.
4. Plugins with zero permissions install immediately without the consent dialog.

Managing Installed Plugins

• Update — when a new version is available, the plugin card shows an update badge. Click to update.
• Uninstall — remove a plugin from your Omniscio installation.
• Settings — some plugins expose configuration in Settings → Plugins.

Ratings & Reviews

Plugin pages show a 1–5 star score and written reviews, which help you judge a plugin before installing it. Leaving your own review needs a marketplace sign-in with GitHub that the regular release of Omniscio does not include yet, so the rating box says so in place of the stars rather than showing a control that could not work. Reading reviews is unaffected.

Security

Every plugin submitted to the marketplace goes through automated security scanning and manual admin review before it's published. The review process checks for:

• Permission abuse or unnecessary access requests
• External URL calls and CDN dependencies
• Obfuscated code or suspicious patterns
• SDK version compatibility

Getting Started

The Omniscio Plugin SDK provides everything you need to build plugins. Install the CLI and scaffold a new project:

npm install -g @agent-mc/plugin-cli
amc-plugin create my-plugin
cd my-plugin && npm install

What You Can Build

Plugins have access to 13 sandboxed APIs across 8 categories:

• Storage & Database — persistent key-value and SQLite storage
• Sessions — listen to session events and status changes
• AI — run AI prompts through the host's credentials
• Filesystem & HTTP — read project files and make network requests
• Cron — schedule recurring background tasks
• UI — sidebar panels, toasts, settings, and custom views
• CLI Endpoints — expose REST endpoints on Omniscio's CLI server
• Events & Logging — structured event system and logging

Developer Dashboard

Access the Developer Dashboard from the Marketplace — click the "Build Your Own Plugins" card at the bottom of the marketplace grid. The dashboard shows:

• SDK documentation links — quick access to the full SDK docs
• Quick-start commands — copy-paste CLI commands to scaffold and publish
• My Submissions — track the status of your marketplace submissions (pending, approved, rejected) with reviewer feedback

Publishing

1. Build and validate your plugin: amc-plugin build && amc-plugin validate
2. Package it: amc-plugin package
3. Authenticate with GitHub: amc-plugin auth
4. Submit to the marketplace: amc-plugin publish

Your submission enters the review queue. An admin will review the code, security scan, and permissions before approving it for the marketplace.

Foundry turns writing a product requirements document (PRD) into a guided conversation. Instead of a blank page, it walks you through a structured, step-by-step AI interview — asking the right questions, drafting each section with you, and assembling a complete spec you can hand straight to a coding session.

What You Can Do

• Guided interview — each step has an AI prompt and a draft you can edit; move forward step by step, or scroll back to rework an earlier answer. Progress saves continuously.
• Autopilot — describe the idea once and let the AI drive every step while you watch a live feed; pause, resume, or abort at any time.
• Red-team reviews (optional, on by default) — adversarial User / Developer / Designer passes that pressure-test the spec before you ship it.
• Export the finished PRD as Markdown.
• Build This Now — once you've exported, hand the PRD straight to a real Claude coding session in a project, pre-loaded with a generated kickoff brief.

RepoGuard is a security checkup for your codebase. It scans a repository for vulnerabilities, exposed secrets, risky dependencies, config problems, and code-quality issues, then rolls everything into a single health score with actionable, fixable findings.

Running a Scan

• Quick Scan — secrets, dependencies, and code quality; fast and low-cost.
• Full Audit — all seven categories: Security/SAST, Secrets, Dependencies, Code Quality, CI/CD & Config, Container Security, and Runtime Health.
• Scan any public repo — paste a GitHub URL into the scan bar to check a repo you don't have locally.

What You Get

• A health score — a weighted 0–100 score and an A–F grade across the seven categories.
• Fleet health — an average across every project you've scanned, plus a table with each project's score, grade, last-scan time, and trend.
• Prioritized findings — drill into a project for category cards and an expandable findings list, each tagged Fix Now / Plan Fix / Improve / Nice to Have (severity × effort).
• One-click fix — the Fix Issues button spawns a fresh Claude session in that project, pre-loaded with the scan report and a prompt to address every finding.
• Export & monitor — copy a Markdown report to your clipboard, and optionally wire up Sentry (free tier) for runtime error trends.

Daily Spend Report posts a single card to your Inbox once a day summarizing your AI spend, built from your live Omniscio ledger. It tells an honest two-part story so you always know what your AI work is actually costing you.

Value vs. out of pocket

The card splits your spend into two very different numbers:

• Value — your agent coding work priced at normal API rates, but covered by your flat subscription. This is the value of the work your agents did, not money you spent. It is usually the big number.
• Out of pocket 🔑 — the real money billed to an API key, i.e. the slice that is not covered by a subscription. This comes mostly from background AI features (like generating session titles). It is usually small, but it is what actually hits your card.

Each card shows both numbers for yesterday, the last 7 days, and the last 30 days, plus a breakdown of yesterday’s spend by AI engine and by background feature.

Turning it on

Daily Spend Report is a plugin, so you install and enable it from the Plugin System (Settings → Plugins). Once enabled it starts posting automatically — there is nothing else to set up.

Choosing the report time

By default the report is posted at 9:00 AM your local time. To change it, open the plugin’s settings and pick a different hour under Report time. If Omniscio wasn’t running at your chosen time, the plugin catches up and posts the report the next time you open the app (once per day, never twice).

The in-app panel

The plugin also adds a Daily Spend Report panel in the sidebar where you can view the same report on demand, any time, without waiting for the daily card.

Omniscio runs several AI agents at the same time, so it needs more memory and processing headroom than a typical desktop app. On an underpowered computer, sessions can run slowly, stutter, or fail under heavy load.

Recommended

• 16 GB of RAM or more — the single biggest factor. 8–12 GB works for light use, but several busy sessions will feel slow.
• A modern multi-core CPU (more than 4 cores).
• Windows 10 or 11 (64-bit), macOS (Apple Silicon or Intel), or Linux (AppImage) — the app ships all three; see the Installation table for the download links.

The low-spec warning

If Omniscio detects that your computer is below these recommendations (under about 14 GB of usable RAM, or 4 or fewer CPU cores), it shows a one-time card in your Inbox explaining what’s low and what helps. The card has one button, Turn on Lite mode, which switches on all of Lite mode’s settings in one click; you can switch it off again any time in Settings → Performance. You can dismiss the card any time, and it won’t keep nagging you.

If your computer is below spec

• Turn on Lite mode (the button on the warning card, or Settings → Performance) — one switch that makes Omniscio itself lighter on a computer like this, including its visual effects.
• Close other memory-heavy apps while running sessions.
• Add RAM if you can — 16 GB or more is the single biggest improvement.

Notice what is not on that list: cutting back your own work. Handling many agents at once is Omniscio’s job, not something you should have to shrink around — if the app cannot keep up on your machine, the switches above are how Omniscio does less, and the rest is ours to fix.

Omniscio supports voice input so you can talk to your AI agents hands-free. This includes a wake word system (like "Hey Siri" but for Omniscio) and continuous dictation.

19

Wake Word

Omniscio can listen for a wake word to start dictation. The wake words are "Computer" and "Hey Omniscio", and both work out of the box. Detection runs fully on your computer, and nothing is recorded or sent anywhere until the wake word fires.

How it works

Wake-word detection uses a built-in free engine: it needs no key and no account, works fully offline, and works out of the box. The two built-in wake words (Computer, Hey Omniscio) come ready to use; their model files are installed automatically the first time you use the engine, with nothing to download and nothing to train. The Settings page shows the exact folder and which wake words have a model installed.

Setting Up

1. Go to Settings → Voice Control.
2. Enable Voice Input.
3. Choose your preferred speech-to-text provider (Deepgram or Groq Whisper).
4. Optionally configure the wake word sensitivity.
5. Click the microphone icon in the toolbar to start listening.

How It Works

1. Say the wake word ("Hey Omniscio").
2. Omniscio plays a chime to confirm it's listening.
3. Speak your message.
4. Omniscio transcribes your speech and types it into the active session's input.
5. The message is sent automatically (or you can review and edit first, depending on settings).

Push-to-Talk

If you prefer not to use the wake word, you can use push-to-talk. Hold a key down to record, release to stop. It works anywhere on your computer, even when Mission Control is in the background.

Push-to-talk is off until you set it up. Go to Settings → Voice Control → Push to talk and turn on one or both keys:

• Push to Command: hold, speak a command, release to run it.
• Push to Dictate: hold, speak, release to type it into the session input.

Pick a key from the suggestions (they are chosen to play nicely with your keyboard layout) or use Advanced (bind any key) to record your own. Very short taps are ignored on purpose, so brushing the key by accident does nothing. On a Mac, the first time you set this up you will be asked to grant Accessibility permission so Mission Control can hear the key while other apps are focused.

Dictation Delivery

After Omniscio captures your speech, it can either send it immediately or place it in the message box for review. Change this in Settings → Voice Control → Dictation delivery:

• Send immediately (default): the transcript is sent to the session the moment transcription finishes.
• Review before sending: the transcript appears in the message box so you can read, edit, and send manually.

If no session is open when you dictate, Omniscio shows a prompt to open one first.

FlowVoice (OS-Wide Dictation)

FlowVoice is Omniscio's OS-wide dictation. Hold a hotkey, speak, release, and your words are cleaned up and typed at the cursor in whatever app is focused: your browser, an editor, a chat window, anywhere you can type. It is separate from the in-app voice features above, which only type into Omniscio's own message box. The transcription service is built into Omniscio, so there are no API keys to set up for it.

Turning it on

1. Go to Settings → Voice Control and find the Dictation (FlowVoice) card.
2. Turn on Enable FlowVoice. The card expands with the full set of controls and three status checks: Server, Microphone, and Accessibility.
3. On a Mac, grant Microphone, Accessibility and Automation access. The card shows the status of each and a Grant access button that opens the right System Settings pane. Accessibility lets FlowVoice react to the hotkey and type at your cursor; Automation (System Settings → Privacy & Security → Automation) is what lets it paste into another app. Grant both — with only Accessibility, dictation still records and still works inside Omniscio, so it can look like it is working while nothing ever reaches the app you are typing in. On Windows, no extra permissions are needed.

Two ways to dictate

• Hold to talk (default): press and hold Ctrl + Shift + D (Cmd + Shift + D on Mac), speak, and release to finish. You can rebind it to a single held key, like the right Cmd key.
• Toggle: press once to start, press again to stop. This second hotkey is unbound by default; record a combo for it in the card if you prefer that style.

Both hotkeys are rebindable right in the card (click the field and press your combo), and you can have both bound at once and use whichever fits the moment.

While you speak

A small floating pill shows what FlowVoice is doing: a live waveform while recording (with accept and cancel buttons), a brief sweep while your words are cleaned up, and a quiet idle pill the rest of the time. Hover the idle pill for quick Dictate / History / Settings shortcuts, or hide it with the Show floating pill toggle; dictation keeps working, and the pill still appears while you are actually dictating. A soft chime marks the start and end of each recording (the Sound cues toggle).

Cleanup levels

Before typing anything, FlowVoice can tidy your transcript. Pick a level in the card: Off keeps the raw transcript word for word, Light strips filler words and false starts, and Strong (the default) fully polishes the wording. You can also pick the language (ten supported) and which microphone to record from.

Dictating into Omniscio itself

When Omniscio is the app in focus, the text lands in Omniscio's message box, and the Send automatically toggle (on by default) submits it for you instead of waiting for Enter. The message box also has its own FlowVoice mic button next to the attach icon: click to start, click again to stop; a small red dot pulses while the mic is live. Right-click that button to open the FlowVoice settings in a dialog right there, without leaving the message you were writing.

History

The settings card keeps a history of your recent dictations: which app the text went to, the cleaned text, and the raw transcript when it differs. You can copy or remove any entry, or clear the whole list.

Voice Commands

Voice commands are separate from dictation. Press Alt + V (the dedicated command hotkey) and speak a command. On Mac the Alt + V chord works everywhere except while your cursor is in a text field, so typing can never falsely trigger it; from a text field use the wake word, the microphone button, or rebind Voice command to a Cmd combo. Omniscio uses AI-powered fuzzy matching, so you don't need to say phrases exactly as listed. A few examples:

• Session navigation — "switch to the marketing session" or "go to session API refactor" jumps directly to a named session.
• Approve inbox item — "approve the request", "accept the request", or "approve the cron job" confirms the currently selected inbox approval.
• Reject inbox item — "reject the request", "decline the cron job", or "deny the request" dismisses the currently selected inbox approval.
• PM: Create item — "create a task called deploy API" or "add an item named review PR" creates a new item on your first board.
• PM: Move item — "move deploy API to done" or "put review PR in backlog" moves an item to a named group.
• PM: Update status — "mark deploy API as stuck" or "set review PR to working on it" updates an item's status column.
• PM: Read item — "read deploy API" or "what is review PR" reads back an item's name and column values.
• PM: List items — "show all items" or "list my tasks" reads back up to 20 items from your first board.

You can also use the Test voice commands button in Settings → Voice Control to try commands without the hotkey.

Voice Report Back

When you ask the AI something with a voice command and it answers in plain prose (for example, "what's on my agenda today?"), Omniscio can read that answer back to you out loud. It uses whichever voice you've picked for Omniscio (Fish Audio, Grok, or ElevenLabs). The feature is off by default. Turn it on at Settings → Voice Control → Voice Report Back.

It only speaks proper written answers from the AI. Quick voice commands that just do something (like snoozing a session, opening your inbox, or approving a request) still show their usual toast confirmation and stay silent. You only hear speech when there's something to actually read out.

If the AI's answer is long, Omniscio speaks the first chunk and then shows a Speak more button on the floating result card. Click it to hear the next chunk. By default the first chunk caps out around 600 characters (roughly 30 seconds of speech), so you don't get a 10-minute monologue from one question. You can change the max chunk size, and set a daily cost cap, under the Advanced area of the same settings panel. The daily cap pauses speak-back for the rest of the day once you've spent that much (ElevenLabs is the priciest voice at about $0.00025 per character, so the default $1 cap buys you roughly 4,000 characters of speech, around 3 minutes; Fish Audio and Grok are much cheaper).

Ask About a Thread (Voice L3)

With this feature you can ask about a specific agent session hands-free, without switching to it or typing anything. Say the wake word (or press Alt+V) and then ask:

• Summary: "tell me about the auth refactor thread" -- Omniscio speaks the Omni briefing for that session.
• Question: "in the auth refactor thread, what is blocking it?" -- Omniscio answers using xAI Grok, reading only that thread's transcript.

After Omniscio speaks the summary or answer, it asks "What would you like to know?" and listens for a spoken follow-up, answers it from the same thread, then asks again -- a real back-and-forth. Each follow-up is answered with memory of the questions and answers already spoken in this exchange, so "what about the second one?" works. The conversation runs until you stay silent, you reach the per-conversation turn limit, or the daily cost cap is hit.

If you do not name a session, or the name does not match any open session, Omniscio falls back to the currently active session.

The feature is off by default. Turn it on at Settings → Voice Control → Ask about a thread (voice). A daily cost cap (default $1) blocks further Grok calls once reached for the day. The operation is strictly read-only -- nothing is queued to the live agent in that session.

Ask the Live Agent (Voice L4)

Voice L4 sends your spoken question to the actual running agent in a thread and tells you when it has answered. Say the wake word and ask, for example: "ask the auth refactor agent why it picked JWT over sessions."

• Omniscio says "Okay, I will ask the auth refactor agent and let you know," and waits for that agent to reach a natural pause before delivering the question.
• When the agent answers, you hear "I have your answer about auth refactor" and get a desktop notification. Click it to hear the full answer.

The question is sent read-only (the agent is asked to answer without changing files), and it appears as a normal message in that thread. The feature is off by default. Turn it on at Settings → Voice Control → Ask the live agent (voice). It only works on threads that are currently running; it never starts a new agent.

Voice answer queue

Ask several running agents at once. Each question shows in a small "Voice answers" panel with a status: Queued, Working, or Ready. When an answer lands, Omniscio waits for a polite moment (no capture, command, TTS, or voice turn in progress) and then announces it. Several answers at once get a rundown ("I have answers on three threads: A, B, and C. Want the A one?") and are read one at a time at your pace:

• "go ahead" / "yes" / "read it" - hears the current answer, then offers the next.
• "next" / "skip" - moves on without reading.
• "what's ready?" (askable any time) - speaks the rundown again.
• silence or "no" - leaves the rest queued; the panel stays visible so you can click to hear or dismiss any time.

The queue is live only and resets when the app restarts.

Spoken output in your language

When Omniscio reads an agent's reply aloud or speaks a Omni briefing, it speaks in the language of the conversation — not always English. The language is detected automatically from the content, so there is nothing to set up. If a voice that matches that language is available, Omniscio uses it for that message; otherwise it keeps your chosen voice, which still speaks the detected language (the built-in voice engines are multilingual). Today all of Omniscio's preset voices are English, so your voice itself does not change — but it already speaks whatever language the content is in, and the matching-voice behavior turns on automatically if a non-English voice is ever added.

Voiceprint Studio

Voiceprint Studio is a separate in-development feature (Settings → Lab → “Voiceprint Studio”) that builds an email style guide in your writing voice — not audio, but how you write. Tag a sample of your sent mail, watch the AI re-draft a real email without seeing your guide, judge its accuracy, and approve fixes that sharpen the guide. The finished guide is something you can paste into any AI to make it write more like you. Voiceprint Studio appears as its own sidebar entry when enabled.

Omniscio's global search lets you find sessions, messages, projects, settings, and more — all from one search bar.

Opening Search

Press Ctrl + K (or Cmd + K on Mac) to open the command palette. You can also click the search bar in the toolbar.

What You Can Search

• Sessions — Find sessions by name, content, or project
• Messages — Search within conversation messages
• Hubs — Jump to a hub
• Settings — Find specific settings by name or keyword
• Actions — Quick actions like "New Session," "Toggle Theme," etc.

Search-as-you-type

You don't have to finish typing a word. The last word of your query auto-matches as a prefix, so Pr finds "Project alpha" and hello wo finds "hello world". Quoted phrases ("hello") and excluded words (-bug) at the end of a query are treated as final — they don't auto-prefix.

Quoted phrases are word-exact

When you wrap text in "…", Omniscio narrows the match to a true word-for-word hit.

• "meeting" finds meeting but not meetings or premeeting.
• Case doesn't matter — "Meeting" and "meeting" give the same results.
• Whitespace inside a phrase is flexible — "project plan" still matches text where the two words are separated by multiple spaces, a tab, or a line break.
• Symbols are literal — "c++ template" finds the exact three-character c++, not "any character."

Search without quotes when you want the broader match: an unquoted meeting still finds meetings and meet via word-stem matching. Use quotes only when you actually want to forbid near-matches.

What gets indexed in a conversation

For Claude conversations the search index is curated so you only land on text that matters:

• Your messages — indexed in full, every word.
• Agent messages — only the final reply after all the tool work. Mid-turn narration, file reads, command runs, and other tool actions are not in the index.
• System rows (the small grey status lines) — not indexed.

So clicking a result always lands you on something a human said or the agent's actual answer — never on Read(file.ts) plumbing or transient mid-turn chatter. Other channels (SMS, Slack, Telegram, RSS, webhooks) are indexed in full because they only carry real message text.

When you don't see results

If a search comes up empty and you have a scope filter applied (the default scope is titles + first message), the empty state shows a "Search all message content for <query>" button. Clicking it widens the search to every word in every message body — useful when your match lives deep in an agent reply rather than a session title.

20

Session tags are short colored labels you attach to a session for quick categorisation — things like urgent, bug, client-acme, or followup. Once sessions are tagged, the Tags virtual hub in the sidebar lets you pick a tag and see every session that carries it, grouped by project.

Tagging is opt-in — turn it on at Settings → Sessions → Session Tags. When the toggle is off, the Tags option disappears from the session menu and the chips vanish from the session header; your existing tags stay in storage and reappear when you turn it back on.

Two kinds of tags

• Free-form tags — short strings you type on the spot in the picker. Quick and one-off; no setup required.
• Library tags — curated entries you create once in Settings → Tags with a fixed name, color, optional description, and project scope. Use these when you want the same label to look and behave consistently across many sessions.

Both kinds coexist on every session and share the same picker, the same chip strip, and a combined cap of 10 tags per session. Tag names are up to 30 characters. Each tag's color is consistent everywhere it appears — free-form tags derive their color automatically from the name, so the same word is always the same color.

Tagging a session

1. Select a session (make sure the message composer isn't focused) and press T — or open the session overflow menu (···) and click Tags…. A centered tag picker opens.
2. Start typing. A list shows matching existing tags (most-used first). Pick one, or — if your text matches nothing — press Enter on the Create "…" row to make a new free-form tag on the spot. Matching is case-insensitive, so typing Bug reuses an existing bug instead of duplicating it.
3. Keep typing to add more tags. Click Browse all… to switch to a checkbox grid of every existing tag with its usage count.
4. Remove a tag by clicking the × on its chip, or press Backspace with the input empty to drop the last one. Press Esc to close.

Tags appear as colored chips in the header strip directly under the session title. (Sidebar session rows don't show chips.) On a phone, when a session has more than one tag the header shows just the first chip plus a +N button to stay compact — tap it to reveal the rest in place, and Show less to collapse them again. Trying to add an 11th tag quietly does nothing — a tooltip explains the 10-tag limit.

The Tags virtual hub

To browse sessions by tag, open the Tags entry in the hub sidebar (pink tag icon, in the Omniscio group between Stats and Settings). It uses the standard three-pane layout: the hub sidebar on the far left, a tag list in the middle column, and the matching sessions on the right.

• Tag list (middle column) — a search box plus two collapsible sections: Library (your curated tags) and Free-form (every distinct typed-in string in use). Each row shows a color dot, the tag name, and a count of how many sessions carry it. A + button in the header adds a new library tag without leaving the view.
• Sessions (right pane) — click a tag and the right pane lists every session carrying it, grouped by project, most-recently-active first. Click any session row to jump straight into it.

Right-click a tag for actions: library tags offer Rename, Change color…, and Delete; free-form tags offer Delete only (to rename a free-form tag, delete it and re-tag with the new word). For library tags, each session row in the right pane also has a small × to remove that tag from just that one session, with an Undo toast.

The view refreshes itself live — tag a session in another window or delete a tag in Settings and the panel updates without a manual reload. Your selected tag and search text are remembered between launches.

Focus Mode collects notifications and delivers them in batches instead of firing one-by-one. This helps you stay in the zone when you're doing deep work.

How It Works

When Focus Mode is enabled, instead of getting a notification every time an agent needs you, Omniscio collects them and sends a single batch notification when a threshold is reached (e.g., "3 items need your attention") or after a time interval.

Configuration

• Count threshold — How many items to accumulate before sending a batch notification
• Time threshold — Maximum time to wait before sending a batch (even if count isn't reached)
• Whitelist — Sessions or projects that always notify immediately (piercing through Focus Mode)

Enabling Focus Mode

Go to Settings → Notifications → Focus Mode and toggle it on. Configure the thresholds to match your preferred interruption level.

Auto-enable during Pomodoro focus blocks

Pomodoro can drive Focus Mode for you. Turn on Settings → Notifications → Alarms → Auto-Enable Focus Mode During Focus Blocks (or set the per-preset override in a Pomodoro preset's Advanced section) and Omniscio will enable Focus Mode when a focus block starts, disable it during breaks (releasing any queued notifications), and re-enable it at the next focus block. When the run ends, Focus Mode is restored to its pre-run state.

Clicking the bell icon yourself mid-run hands control back to you for the rest of that run — Omniscio will not flip Focus Mode on the next phase change. The next run starts fresh with automatic management.

Not the same as the Shift+F filter

Shift + F switches a different filter with a similar name: it makes a hub’s session list show only the sessions that need you, hiding running, paused and interrupted ones. The batch-alerts Focus Mode above never hides a session. While that filter is on, a banner pinned at the top of the sidebar says focus mode is hiding sessions, with a Show all button that turns the filter off, and the Live Sessions heading still counts your running sessions. Press Shift + F again, or click Show all, to see everything.

This filter used to answer to a bare F — one stray keypress away from emptying a whole hub, with the only hint a quiet line that scrolled away with the list. It takes a deliberate chord now, and the banner above the list keeps saying so for as long as it is on.

Presentation Mode makes Omniscio safe to screen-share, record, or stream. Flip it on and it hides your real projects and sessions, your account name and email, live cost and spend figures, and the leaky bits of header chrome — revealing only the items you explicitly pick, plus any session you start yourself while it's on. Nothing is renamed or changed under the hood; it only affects what's drawn on screen, so turning it off restores everything instantly, and it stays on across an app restart.

Turn it on from the Presentation Mode item in the header’s “…” overflow menu (pin it to the header and it lights up with a “LIVE” badge while active), or the command palette. There is also a “hide everything now” panic control that instantly hides everything, including anything you’d revealed — it has no keyboard shortcut out of the box, so give it one in Settings → Keyboard Shortcuts if you want one. (It used to ship on Ctrl + Shift + X and could fire while you were typing, which made it far too easy to trigger by accident.) You choose what shows from the backstage staging panel, where you also pick a hide style for everything else:

• Remove — the item is filtered out of the list entirely (the default, and the safest).
• Blur — the item stays in place but its name is blurred.
• Placeholder — the name is replaced with a neutral label like “Project 1” or one you type in.

If it ends up hiding every project, the sidebar says so in plain words and offers a one-click “Turn off presentation mode” button, so a blank list is never a dead end. It deliberately shows no project names and no counts there — that screen is the one on camera.

While presenting, routine pop-up notifications, the app-icon badge count, and phone push are all held so nothing flashes on camera (real errors still show). You can set a stand-in demo name and avatar in place of your real account, optionally blur the message content inside an open session, and — if you don't want the header account switcher on camera at all — flip “Hide the account widget” to remove that pill (and its account-balancer count) from the header entirely while presenting.

If Remove ends up hiding every project, the sidebar tells you so rather than going blank — it says Presentation Mode is on, confirms nothing has been deleted, and gives you a one-click button to turn the mode back off. Your phone does the same. If only some are hidden you get the ordinary “Showing 3 of 12” strip instead, and its reveal button turns off whichever thing is actually doing the hiding. The empty message deliberately shows no count and no names, because that screen is the one on camera.

Two optional switches: a quiet inbox and paused automatic sessions

Both are off by default, and both only do anything while Presentation Mode is on — so turning the mode off restores everything with no second switch to remember. You’ll find them in the staging panel.

Keep the inbox quiet empties your inbox on camera — every source at once: sessions that need you, agent alerts, email, SMS, Telegram, GitHub pull requests, Jira, approvals, digests and plugin rows. The inbox badge goes to zero with it. Nothing is deleted, archived or delayed: every row still exists the whole time and comes straight back the moment you turn the switch (or the mode) off. While it’s on, every inbox screen says why it’s quiet and offers a one-click “Show my inbox”, so a quiet inbox is never mistaken for “all clear”.

Pause automatic sessions stops Omniscio starting new work while you present — schedules, automations, away-mode, the auto-lander, inbound email, bug intake, and sessions other agents ask for. Two things it deliberately does not do:

• It never blocks you. Anything you launch yourself — “+ New Session”, a quick-launch button — starts normally.
• It doesn’t throw the work away. Sessions requested from the command line or by another agent are genuinely queued and start the moment you exit. Scheduled jobs are skipped rather than failed, so they run on their next tick with nothing recorded against them.

If Omniscio restarts while you’re still presenting, your own interrupted sessions come back as usual — only new automatic work keeps waiting. That way a crash mid-presentation can never leave your real work parked behind a switch you forgot to turn off.

Bookmarks are a hierarchical, browser-bookmark-bar-style launcher that lives in the Omniscio toolbar. You can save URLs, file or folder paths, programs, shell commands, and shortcuts that jump to a specific session, project, or virtual hub inside Omniscio. Bookmarks are opt-in — flip them on at Settings → Widgets → Bookmarks → Enable bookmarks (the toolbar bookmark icon appears the moment you do).

Everything is local-only. Bookmarks live in Omniscio's SQLite database on your machine and never sync anywhere. The only outbound network call the feature makes is a best-effort favicon lookup so URL rows show a recognisable icon.

Opening the popover

Click the small bookmark icon in the Omniscio toolbar pill. A popover hangs from the toolbar with a search box at the top, a small + button for adding new bookmarks, and your hierarchical tree below. Press Esc, click anywhere outside the popover, or click the toolbar icon again to close it.

The eight kinds

Each bookmark has a kind that controls its icon and what happens when you click it. Five of them launch something external; three of them navigate inside Omniscio.

External launchers (five)

• Folder — a container for other bookmarks. Click the row to expand or collapse it. Folders can nest up to 5 levels deep.
• URL — an http:// or https:// link. Click opens it in your default browser. The row icon is the site's favicon when Omniscio can fetch one, otherwise a small globe.
• Path — a file or folder on disk. Click opens it in your OS file manager (Explorer / Finder) or in the default app for that file type. Environment variables in the %NAME% form are expanded.
• Executable — an absolute path to a program. Click launches it in the background (no console window).
• Command — an arbitrary shell command. Click runs it through cmd.exe /c. The first time you click a command bookmark, Omniscio pops a confirmation modal showing the literal command and asks you to tick "I understand" — only then will the command fire. After that one confirmation, subsequent clicks run silently. Editing a command's value re-arms the gate.

In-app navigation (three)

These bookmarks don't open anything external — they make Omniscio jump to a target inside the running app.

• Session — pinned to a specific session. Click jumps to that session (activating its parent project first if needed).
• Project — pinned to a specific project. Click activates that project in the sidebar.
• Virtual — pinned to a virtual hub (Inbox, Gmail, Claude, Briefings, Stats, etc.). Click activates that virtual hub.

Nav bookmarks store the target's underlying ID, not its name — so renaming a session or project in the sidebar updates the matching bookmark's label automatically the next time you open the popover. If a target is deleted, the row gets a strike-through and an amber warning icon, and clicking it pops a toast with a one-click Remove bookmark action.

Adding a bookmark

1. Open the popover and click the small + button in its header (or, if you have no bookmarks yet, the Add bookmark button in the empty state).
2. Pick a kind at the top of the modal: Folder / URL / Path / Executable / Command / Session / Project / Virtual.
3. Fill in the rest of the form. For URL / Path / Executable / Command, type the value freehand. For Session / Project / Virtual, the value field is replaced by a target picker — type to filter the list of live sessions, projects, or virtuals in your Omniscio install, then click one to select it.
4. Optionally pick a parent folder from the dropdown (or use + New folder… at the top of the dropdown to create one inline without leaving the modal).
5. Click Save. The popover's tree refreshes with the new row.

Searching, reordering, and renaming

• Search — type in the box at the top of the popover. The filter matches the bookmark's name or its value (so you can find a URL row by typing part of the URL even if it has no name). Matching rows and their parent folders stay visible; everything else hides.
• Reorder — drag a row up or down. Drop on the upper or lower edge of a row to reorder, or drop on the middle of a folder row to move the row into that folder. Hovering over a collapsed folder for ~0.6s auto-expands it so you can drill in mid-drag.
• Keyboard reorder — focus a row, then Alt+ArrowUp / Alt+ArrowDown swaps it with its sibling.
• Rename a folder inline — double-click the folder name in the tree.
• Edit any row — right-click it to open the same modal you used to create it. You can change the name, value, or parent folder. (Switching a row's kind isn't allowed — delete and recreate instead.)
• Delete — right-click → Delete. Deleting a folder asks for confirmation and removes everything inside it.
• Middle-click a non-folder row to launch it without closing the popover — handy for opening several URLs in a row.

Settings

• Settings → Widgets → Bookmarks → Enable bookmarks — the master on/off toggle. OFF by default. When off, the toolbar icon disappears and the popover is unreachable; your saved rows are not deleted, just hidden.
• Settings → CLI Control → Enable bookmarks CLI — an independent kill switch for the external HTTP control of bookmarks. Most users won't touch this; it only matters if you're scripting Omniscio from outside.

Quick Launch is a global-hotkey floating composer that pops up over whatever you're doing — think Spotlight on macOS or Alfred on Windows. Type a prompt, hit Send, and a brand-new Claude session starts running it — and if you launched Quick Launch from another app, it leaves you right there instead of pulling you into Omniscio (see below).

Open the modal

Press Ctrl+Space (the default) from anywhere on your computer. A small centered window appears over your current app with a textarea, a project chip, a Send button, and — only when voice input is enabled (off by default) — a microphone button.

If the modal is already up and focused, the second hotkey press is a no-op. If it's up but not focused, the second press refocuses it and selects all the text so you can either keep typing or just start over.

Type and send

• Enter submits; Shift+Enter inserts a newline. This follows your global Submit Key setting (Settings → Sessions) — switch it to Ctrl+Enter to send and Ctrl+Enter (or Cmd+Enter) submits while Enter makes newlines.
• Esc — or Ctrl+W (Cmd+W on mac) — clears the draft and hides the modal.
• Click outside the modal to dismiss it.

On submit, a new Claude session is created in the selected project and your prompt is sent as the first message. If you opened Quick Launch from inside Omniscio, its window comes to the front so you land on the new session; if you opened it from another app, you stay in that app — Quick Launch won't pull you away.

Quick-add tabs

A thin tab strip across the top of the modal lets you swap the body for a different "quick-add" mini-form without leaving Quick Launch. The default pinned tabs are Session · Search · Ask Omniscio · Automation · Alarm · Task · Calendar — Session first, then the three tabs that start a session, so the strip opens on what you actually came to do. The last three are only visible once their feature is on (Alarm needs Alarms, Task needs Tasks, Calendar needs Google Calendar connected), so a brand-new install shows exactly four: Session, Search, Ask Omniscio and Automation. Drip and Bookmark are no longer pinned by default — they are still full tabs, one click away from the strip's + button. If your strip is still the untouched old default, Omniscio moves you onto the new one once, automatically; if you have ever reordered, removed, or added a tab, your strip is left exactly as it is. Email, SMS, Vault Note, Find Note, and Find Email ship in the registry but default to unpinned — pin them from the Available column at Settings → Quick Launch to surface them in the strip (the Email tab additionally needs Quick Email enabled — see Email — send a quick note in two keystrokes below):

• Alarm — text-only single textarea. Type a phrase like "every weekday at 7:30 AM stand-up", "tomorrow at 9am", or "first Monday of the month at 9am", hit Enter, and Omniscio schedules the alarm. The same parser the Calendar tab uses handles both the legacy 5-shape recurrence (once / daily / weekdays / weekends / specific days-of-week) and a richer RRULE escape hatch for things the 5-shape list can't express ("every 3 weeks", "first Monday of the month", "every other Tuesday"). A preview line above the textarea shows the inferred schedule as you type. Most parses are free (offline regex); ambiguous prose falls back to a Claude Haiku call capped at a small daily budget shared with the Calendar tab. The alarm is created locally — no Google account needed; it rings inside Omniscio like any alarm from the Alarms virtual hub. Hidden when Alarms are disabled at Settings → Notifications → Alarms.
• Task — quick-add a to-do without opening the Tasks screen. Type what you need to do, press Enter, and it lands in your Tasks Inbox. Need it at a certain time? Open the small Schedule section for three one-tap presets (This evening, Tomorrow morning, Next week) or an exact date and time, plus a Remind / Hide-until choice: Remind keeps the task in your list and drops a reminder in your Inbox at that time; Hide until tucks it away until then, when it pops back with a reminder. Leave the Schedule closed and it’s just a quick capture, exactly as before. Works the same on your phone’s quick-actions sheet.
• Calendar — type an event in plain English in a single Event field ("Lunch tomorrow at noon", "Standup every weekday 9am", "Birthday tomorrow 10am yearly") and Omniscio creates the event on your Google Calendar. Type a time zone with the time ("9am EST", "2pm PST", "noon eastern time") and it's honored — the event is booked at that clock in that zone (daylight-saving aware, so summer "EST" is Eastern Daylight Time), not silently shifted to your computer's zone; leave the zone off and it uses your local zone as before. By default the event lands on your primary calendar, but if you have more than one calendar you can write to, a calendar picker appears below the preview and remembers your choice for next time. Recurrence is extracted from the same input — including a dangling “…every” typed after the day (“Kathleen Tuesday 10am every” → repeats weekly on Tuesday). When the text alone doesn't imply a repeat, a small Repeat checkbox + Daily/Weekly/Monthly/Yearly picker below the preview lets you force one; it mirrors any repeat the text already implied, so it never disagrees with the preview. Each preview row reads as Title — date and time spelled out in full (e.g. Monday, June 29 at 12:00 PM), with the date/time hugging the title rather than pinned to the right edge; an all-day event shows the date with no time. If you gave an exact time — including compact forms like 805p or a range like 8-9p that sets start and end (a range still works with a day-of-week trailing it, e.g. "block 2-4pm thursday") — the preview is a single timed row; if you gave a date only ("Friday birthday"), you get two rows — an all-day interpretation and a 9 AM timed alternative — so you can pick which one you meant. This tab shares its date/time engine with session snooze, so anything the Snooze box understands (business days, ordinal days, holidays, bare months, and more — see Snoozing Sessions) works here too. The offline regex parser handles ~99% of inputs for free; the rare miss falls back to a Claude Haiku call (capped at a small daily budget shared with the Alarm tab) and shows a small "AI is thinking…" sparkle indicator so you know that particular parse took longer because your phrasing was unusual. After a successful create, the composer closes and a small toast appears in the bottom-left of the main Omniscio window — “Added to Google Calendar” — with Open and one-click Undo links. Visible only when Google Calendar is connected at Settings → Google Workspace.
• Search — type a query, hit Enter, and Omniscio spawns a new session in the Session Search project that uses Claude (with the bundled search-sessions skill) to find matching sessions across your history. Accepts the same attachments as the Session tab — paperclip / paste / drag images, PDFs, or modern Office (.docx/.xlsx/.pptx) and plain text files. Attachment-only submits are allowed: leave the query blank, drop a screenshot of an error, hit Enter, and Claude will mine the image for search terms (error text, code, UI labels, filenames) before running the search.
• Ask Omniscio — type a question about Omniscio itself ("how does X work?", "where do I configure Y?", "why is Z happening?"), hit Enter, and Omniscio spawns a new session in the Ask Omniscio project preloaded with the codebase context Claude needs to answer.
• Email — fire off a short email in two keystrokes: add one or more recipients, type a line, press Ctrl+Enter, and it sends through your connected Gmail account after a brief undo window. Off by default — enable it and add recipients at Settings → Email & Summaries → Quick Email. Full walkthrough below under Email — send a quick note in two keystrokes.
• SMS — fire off a text the same way: pick someone you’ve texted (or a saved favorite, or type any number), write a line, press Ctrl+Enter, and it sends through your SMS connection after a brief undo window. Off by default — enable it at Settings → Channels → SMS. Full walkthrough below under SMS — send a quick text.
• Vault Note — jot a quick note straight into your The Vault vault without opening the Vault screen. Type in the single text box, press Enter (Shift+Enter for a new line), and the note is saved and the modal closes. The first line becomes the note’s title plus a timestamp (e.g. Buy milk 2026-06-08 1430); it lands in an auto-created Quick Notes folder in your active vault (the one The Vault is currently pointed at), and the full text is the body. Two notes saved in the same minute never overwrite each other. Visible only when The Vault is turned on at Settings → Features.
• Find Note — a fast "jump to any The Vault note" switcher (Omniscio’s take on the standalone Nothari app’s Find tab). Start typing a note’s title and it fuzzy-matches as you go; an empty box lists your most-recently-edited notes for a one-keystroke jump back. Arrow keys move the highlight, Enter opens the note in the main window’s Vault panel (Omniscio foregrounds itself — even if the Vault screen wasn’t open yet), Esc closes the modal. Unlike the other tabs it only navigates — it never creates anything. It also has its own dedicated global hotkey (default Ctrl+Shift+Space) that opens Quick Launch straight to this tab from anywhere; rebind or disable it at Settings → Keyboard Shortcuts (the "The Vault quick find" row). Visible only when The Vault is turned on at Settings → Features.
• Find Email — a fast search-your-email tab: type a query and Omniscio searches your Gmail, listing matching emails right there with the sender, subject, a snippet, and the date. Arrow keys move the highlight, Enter (or a click) opens the highlighted email — the main window comes forward, switches to your Gmail view, and the email opens in Omniscio’s built-in reader; Esc closes the modal. If you use Supermail, turning it on switches this tab to search Supermail and open the picked email right in Supermail instead — otherwise it searches Gmail exactly as described. Like Find Note it only navigates — it never creates or sends anything. An empty box shows a prompt rather than loading anything, since each search checks Gmail live, so nothing runs until you type. Visible only when Gmail is turned on at Settings → Email & Summaries → Gmail; unpinned by default.
• Automation — describe an automation in plain words in a single text box ("email me a summary of my unread mail every morning at 8"), press Enter (Shift+Enter for a new line), and Omniscio starts an Automation Helper session seeded with your description — the same "describe your own" flow as the Automation Helper page, one keystroke from anywhere. The Helper handles the technical work for you — it asks any follow-up questions it needs, builds and tests the automation, and schedules it — and nothing runs until you approve it. Visible only when the Automation Builder is turned on at Settings → Features → Enable Automation Builder.
• Command Palette — a searchable list of Omniscio’s main actions, so you can start any of them straight from Quick Launch. Type to filter (by name or a synonym — "diagram" finds Flowchart), arrow keys move the highlight, Enter (or a click) runs it — New Mind Map, New Whiteboard, New Flowchart, message a teammate, New Note / Task / Bookmark / Alarm, open the Browser or the Agent Status Board, and more. Actions for features you’ve turned off don’t appear, so the list only shows what you can use. Because Quick Launch is its own window, picking an action brings Omniscio to the front and does it there. Desktop-only; unpinned by default.
• Settings — search your Omniscio settings and change the simple ones right in the pop-up instead of opening the full Settings screen. Type to filter; a toggle, dropdown, number, or short-text setting shows its control inline and saves the moment you change it, while a more complex one (account connections, saved keys, keyboard shortcuts) shows an "Open in Settings →" link that jumps you straight to that setting in the full Settings screen, highlighted. Desktop-only; unpinned by default.

Press Ctrl+Tab to cycle forward, Ctrl+Shift+Tab to cycle back, or Ctrl+1–Ctrl+9 to jump straight to the 1st–9th tab (hover a tab to see its number). Submitting a non-Session tab (press Enter; Shift+Enter makes a newline and Ctrl/Cmd+Enter still works — except the Email tab, whose multi-line body sends on Ctrl/Cmd+Enter so a plain Enter starts a new line) saves the entry through the same backend the main app uses and dismisses the modal.

At the right of the strip, a small + button lets you add a tab on the spot — click it for a short menu of the tabs you have available but haven’t pinned yet (like Search or Ask Omniscio), and pick one to add it instantly. Next to it, the gear icon hides Quick Launch, brings Omniscio to the front, and jumps straight to Settings → Quick Launch — a two-column card where you can pin/unpin tabs or drag-handle to reorder them. You can also drag the tabs right in the strip — grab any tab (Session included) and drop it in a new spot to reorder; the order saves instantly and a quick tap still just switches tabs. When you pin more tabs than the fixed-width window can show, the tab row scrolls sideways instead of pushing anything off the edge — a round ◀ / ▶ arrow fades in on whichever side has more tabs (click it, spin the mouse wheel over the tabs, or press Ctrl+Tab to scroll along), while the + and gear stay pinned at the right and are never clipped. Session is always pinned — you can’t remove it — but you can drag it anywhere. Unpin every other tab and the strip collapses to the single Session pane — unless you still have a tab available to add, in which case the strip stays so the + is one click away.

Pick a project

The chip row in the modal shows the active project with a small down-caret on the right so the chip reads as expandable. Click it (or press Ctrl+P) to open a narrow fuzzy-search picker anchored to the bottom-left of the modal — type to filter, arrow keys to navigate, Enter to pick. Projects without a custom icon show a colored initial-letter glyph so every row stays visually distinct. Long names ellipsize so the picker stays narrow and the rest of the modal remains click-able for dismissal.

Attach files

• Paste an image from the clipboard.
• Drag a file onto the modal.
• Click an image chip to preview it full-screen — zoom, arrow-key between images, copy.
• Right-click an image chip for Copy / Save Image As… / Remove.
• Drag a chip to reorder; middle-click it or click its × to remove.

The attachment travels with the prompt to the new session. Press Esc to close a preview — or any open menu, dropdown or popover — without closing Quick Launch; a second Esc closes the window.

Dictate (voice)

The microphone button appears only when voice input is enabled (Settings → Voice Control) — it's off by default, so out of the box there is no mic here. With voice on, click the microphone button to dictate the prompt by voice. While the Quick Launch modal has focus, Omniscio suppresses its main-window voice command parser — so transcripts get appended to the modal's textarea instead of triggering voice commands. Saying "create a session in the trading project" puts that literal text in the textarea, ready to edit and send.

On your phone

Quick Launch's pop-up window is desktop-only, but the same actions are one tap away on mobile. Tap the ⚡ button in the top bar and a sheet slides up with your pinned quick actions as tiles — tap one (Alarm, Task, Bookmark, Calendar…) to fill in its little form right there, or tap New Session to start a session on the spot. It shows exactly the tabs you've pinned on desktop, so there's nothing separate to set up. On the single-line tabs (Search, Bookmark, Web Clip) your phone keyboard's Enter/Go key submits the form — and on the find-and-pick tabs (Find Note, Find Email, Super Prompts) it opens the highlighted result — so you don't have to reach for the button.

Settings

All four Quick Launch settings live at Settings → Quick Launch:

• Quick Launch Hotkey — the keyboard combination that opens the modal. Default Ctrl+Space. Click the X to disable. If Ctrl+Space conflicts with your IME or another global tool, rebind it (e.g. Ctrl+Alt+Space).
• Enable Quick Launch — on/off toggle (default on). When off, the hotkey is ignored.
• Quick Launch Default Project — which project new sessions start in. Default Last used project remembers your last pick; you can also pin a specific project.
• Quick Launch Tabs — two-column pinned/available card for picking which quick-add tabs show in the modal's tab strip and in what order. Session is always pinned but can be reordered anywhere — you just can’t remove it; the other six default-pinned tabs (Search, Ask Omniscio, Automation, Alarm, Task, Calendar) can be reordered or removed. You can also right-click any tab in the strip for a quick menu — a Global hotkey field to give that tab its own shortcut on the spot, plus (on every tab except Session) a Remove option to unpin it — all without opening Settings. Quick Launch opens on whatever tab you put first. The Calendar tab only shows in the strip when Google Calendar is connected, and the Task tab only once Tasks is on. Drip, Bookmark, and Email sit unpinned in the Available column — check their boxes to add them to the strip. A row whose feature is off is grayed out and can’t be checked until you turn that feature on (the Email tab, for instance, needs Quick Email on first) — enable it, then pin it. A tab you’d already pinned before its feature was turned off stays in the Pinned column with a short note next to it explaining why it isn’t in the strip yet. Reachable directly from the modal's gear icon. A Tab hotkeys sub-section at the bottom lets you give any tab its own global shortcut (e.g. Ctrl+Alt+C) that opens Quick Launch straight to that tab from anywhere — even a tab you haven't pinned. It needs a modifier (Ctrl/Alt/Shift), is gated by the main Enable Quick Launch toggle, and a combo that clashes with an existing Omniscio shortcut is safely skipped. You can set the very same shortcut right on the tab via its right-click menu, and project tabs can take one too (their shortcut opens a new session straight into that project).

Email — send a quick note in two keystrokes

The Email tab turns Quick Launch into a speed-dial for short emails: add one or more people, type a line, press Ctrl+Enter, and it sends through your connected Gmail account. It’s for the one-liners you fire off constantly (“running 5 late”, “approved, go ahead”) — not a full mail client. You can put several people on one email’s To line (up to 10 — a quick note, not a mailing list; everyone is on To and can see each other), and an Add Cc/Bcc link adds optional Cc and Bcc fields (each also up to 10, all on the same one email; Bcc stays hidden as usual). There’s no thread view (though you can attach files and paste screenshots — see below); the subject line is optional and hidden by default (leave it blank and the email goes out with a single hyphen — - — as the subject line).

Email is off by default. Turn it on at Settings → Email & Summaries → Quick Email, then pin the Email tab from the Available column at Settings → Quick Launch. Once Quick Email is on, a pinned Email tab shows in the strip even before you’ve added a recipient — it opens to an add-a-recipient screen that points you to Quick Email settings. (If a tab you pinned isn’t showing, the Quick Launch tabs editor shows a short note next to it explaining why.) Sending uses the same Gmail account Omniscio already uses for the Gmail integration and Email Summarizer. And once Quick Email is on, pressing Alt+E from anywhere opens Quick Launch straight to the Email tab — a built-in default you can rebind or clear alongside the other Quick Launch tab shortcuts.

The tab has a Recipient chip field (it starts empty — whenever you open the tab without a specific person already chosen, e.g. with Alt+E, the cursor starts right here so you pick who first, dropping the people list open). Add as many people as you like — up to 10, each shown as a removable chip, all on one email’s To line: click a suggestion or press Enter/Tab to add one, and Backspace on the empty field removes the last chip. It suggests your contacts and the people you email with — start typing a name or email and Omniscio surfaces matches from your Google Contacts, the people you email most, and the people who email you (so someone you’ve only received mail from still turns up). As you type it also runs a quick live search of your Google Contacts to catch anyone not already in the list. Automated no-reply senders are filtered out, the people you actually email rank first, and the list keeps up with your sending — email someone new and they show up the next time you open the picker. This needs no extra Google permission beyond what Omniscio already uses. Pick suggestions or type any email addresses to send to anyone. Need to copy people? An Add Cc/Bcc link reveals matching Cc and Bcc chip fields that work exactly the same way (up to 10 each, all on the one email; Cc is visible to everyone, Bcc is hidden); you still need at least one To recipient, and a send with no Cc/Bcc goes out just as before. An optional Add subject link and a Body box (which takes the cursor on open instead when a per-recipient hotkey already chose the person) with emoji input — type : then two characters for autocomplete or click the smiley button for a full picker — complete the composer. The Email composer opens taller than the other Quick Launch tabs so a longer message has room to breathe, and grows a touch more when the subject is showing. Two shortcuts jump you straight to a field from anywhere in the composer: Ctrl + Shift + S for the Subject (revealing it first if it’s hidden) and Ctrl + Shift + M for the Message. Send stays disabled until you’ve added at least one recipient and a subject, a body, or an attachment — a subject-only note, or an attachment with no words, is fine; only a completely blank send is blocked. Press Ctrl+Enter (or click Send email) and the composer closes immediately so you’re back to what you were doing — a plain Enter just starts a new line in the body (and from the subject line jumps the cursor down to the body), so you can write a few lines before you send.

Attach files & screenshots. A quick email can carry attachments three ways: paste a screenshot (Ctrl+V) straight into the message box, click the paperclip in the footer to pick files (images, PDFs, Office docs, text), or drag files onto the composer. Each shows as a small removable chip (a thumbnail for images); unsupported or oversized files are skipped with a short note, and the total is capped around 20 MB (Gmail’s own limit). A screenshot with no words is a valid send. Staged attachments are kept with your draft and clear when you send.

The undo window

Your email doesn’t leave right away. The instant you send, the composer closes and a “Sending to <name>…” toast appears in the main Omniscio window with an Undo button (for several recipients it reads as a short summary, e.g. “Sending to Alice +2…”). The message is held briefly — 10 seconds by default, adjustable anywhere from 1 to 30 seconds at Settings → Email & Summaries → Quick Email. Cancel during that window by clicking Undo or pressing Ctrl+Z. When the window passes with no undo, the email actually sends and the toast becomes a quick “Sent to <name>” confirmation (or an error toast if Gmail rejected it).

Undo brings your email back. When you undo a send, the composer reopens with the whole message restored — recipients, subject, body, and attachments — ready to tweak or resend, just like Gmail’s undo. A failed send keeps your email too, so it’s waiting for you to retry the next time you open the composer. The one exception: if you’d already started a new email during the undo window, that one is kept and the undone message won’t overwrite it.

Ctrl+Z is global while a send is pending. Omniscio grabs Ctrl+Z system-wide for those few seconds so you can undo from any app without clicking back into Omniscio, and it undoes the most recent pending send first. The moment nothing is pending, Omniscio lets go of Ctrl+Z — so it only ever shadows undo in your other apps during a live send window, never permanently.

Send later

Need it to go out later instead of now? Press Ctrl+Shift+L — or click the clock button in the footer — to open a small Send later picker. Choose a preset (“2 hours from now”, “tomorrow morning”, “next week”…) or type a time in plain words (“tomorrow 9am”, “friday 3pm”), and the composer closes with a “Scheduled for <time> · Cancel” toast. A scheduled email is saved and delivered at your chosen time even if you close and reopen Omniscio in between; click Cancel on the toast to call it off before it goes. Scheduling runs inside Omniscio, so a send fires while the app is running — if your computer was off at the exact moment, it goes out as soon as you’re back.

Your draft is saved

If you close the popup without sending — Escape, click away, or Cancel — your half-written email is kept, just like the regular new-session box: reopen the Email tab and your recipients, subject, and message are all still there. It follows the same Settings → Quick Launch → “Remember unsent Quick Launch text” switch (on by default); turn it off and the Email tab starts blank after every close. (Bringing the recipients back is just your own earlier pick returning — Omniscio still never auto-addresses a quick email to a default person; and a recipient whose saved entry you’d deleted meanwhile comes back as its plain email chip — the exact address you chose, never re-pointed at someone else.)

To throw a draft away on purpose, click the small eraser in the top-right corner of the message box (it only shows once there’s something to clear). It wipes the whole email and shows a “Draft cleared · Undo” pill — click Undo (or press Ctrl+Z) to get it back. Sending clears the draft too. The draft is remembered in memory only, so a full app restart starts blank; and undoing a send does not bring its draft back.

Per-recipient hotkeys

Each saved recipient can carry its own global hotkey that opens Quick Launch straight to the Email tab, pre-addressed to that person with the cursor in the message body (the “who” is already settled) — so a hotkey assigned to your manager pops a blank quick-note to them from anywhere, even when Omniscio is in the background. These are off until you flip the Recipient hotkeys switch and assign a hotkey to a recipient. Omniscio’s built-in shortcuts (focus app, Omni briefing, Quick Launch) always win a conflict.

Quick Email settings

Everything is at Settings → Email & Summaries → Quick Email:

• Enable Quick Email — the master on/off (default off). Turning it off hides the Email tab and disables every recipient hotkey.
• Recipients — up to 50 saved people, each with an Email address (required), an optional display Name, and an optional hotkey. Click Add Recipient, fill in the row, and it saves automatically when you click away — these are the rows the composer dropdown shows. Drag a row by its grip handle to reorder; this order is the dropdown order, so put your most-used person on top for the quickest pick (the composer’s To field starts blank, so this sets the list order, not a default recipient). Or use the Search your contacts box at the top to pick from your real people — Omniscio pulls from your Google Contacts address book (for names) and your recent sent mail (for frequency ranking), merged into one list with your most-emailed people first. Existing users whose Google connection predates this feature see a one-time Reconnect Google inbox nudge to grant the read-only Contacts permission; new users get it automatically. If contacts can’t be read, you still get sent-mail suggestions — and you can always add recipients by hand.
• Recipient hotkeys — the master switch for the per-recipient global hotkeys above (default off).
• Undo window — how many seconds a message is held before it sends (default 10, anywhere from 1 to 30).
• Show subject by default — whether the Subject field is visible when the tab opens (default off — body-only is the common case).
• Signature — an optional short sign-off added to the bottom of every quick email you send, after a blank line (default empty). Type it once in the box and it’s saved when you click away. It’s separate from your Gmail signature — quick emails go out through Gmail’s back end, which never adds your Gmail signature, so this is how you get one on them. Leave it blank and your quick emails stay bare, exactly as before.

Forward Email autocomplete

The same contact suggestions also appear when configuring a Forward Email automation action. The recipient field in the Forward Email config offers autocomplete from the same sources — your Google Contacts, the people you email, and the people who email you — so you get the same people suggestions when setting up forwarding rules as you do in the Quick Email composer. As you type (2+ letters) it also runs the same live search of your Google Contacts, merged below the instant matches and skipping anyone you've already added. Type a name or email to search, press Tab or Enter to accept a suggestion, and Backspace to remove a recipient chip.

SMS — send a quick text

The SMS tab is the text-message sibling of the Email tab: pick a recipient, type a line, press Ctrl+Enter, and it sends through your SMS connection. It’s for the quick texts you fire off constantly — not a full messaging app (no pictures, GIFs, or conversation threads; those live in the main SMS view).

SMS is off by default and needs your SMS connection set up. Turn it on at Settings → Channels → SMS (connect Pushbullet or your own phone endpoint), then pin the SMS tab from the Available column at Settings → Quick Launch. If SMS is on but not connected yet, the tab tells you so and points you to Settings instead of showing a form that couldn’t send.

The To box merges your saved favorites with people you’ve recently texted, and — because a phone number is an open set — lets you type any number to text anyone (a “Text <number>” row appears once what you type looks like a number). Type your message and press Ctrl+Enter (or click Send text); a plain Enter just starts a new line. The composer closes and a “Sending to <name>…” toast appears in the main window with an Undo button for a few seconds — click Undo and nothing is sent; let it pass and the text goes out (or you get a plain-language error if it couldn’t).

SMS Favorites

Under Settings → Channels → SMS there’s an SMS Favorites card — pin the people you text most (a name is just a label; only the number is texted) so they’re one pick away. Click Add Favorite, fill in the row, and it saves when you click away once the number looks valid. You don’t have to build this list by hand — everyone you’ve already texted shows up in the tab automatically; favorites just pin the ones you use most.

Click Browser in the Omniscio sidebar (globe icon) and the right pane becomes a full browser: tabs, bookmarks (star a page, manage them from the bar under the toolbar), history, find-in-page (Ctrl+F), zoom, and the usual shortcuts — Ctrl+T new tab, Ctrl+W close, Ctrl+L address bar, Ctrl+R reload. Links that try to open a popup or a new window open as a new in-app tab instead. Downloads land on a shelf at the bottom with progress and “show in folder”; a file type that can run programs asks you before it saves. If Browser isn’t in your sidebar, turn it on at Settings → Features → Enable Browser.

What makes it the AI Browser: an Omniscio session can drive it too — open a page, read it, click, type, take a screenshot — but only in its own separate tabs, on its own profile, walled off from yours. The agent never sees your cookies, your logins, or your saved passwords, and a layered safety model (below) decides what it may do and when.

Tabs and Spaces

• Spaces — named groups of tabs (“Work”, “Research”). Switching spaces swaps which tabs you see. They’re tidy piles, not separate logins — every space shares the same sign-ins.
• Vertical or horizontal tabs — a toolbar button swaps the tab strip from across the top to down the side. Your choice is remembered.
• Pin / Favorite / Today — right-click a tab: Pin keeps it alive indefinitely, Favorite protects it from auto-archiving, and everything else is Today — eligible for Omniscio’s automatic tidy-up of tabs you haven’t touched in a while. The tab you’re on (and the agent’s own tab) is never auto-cleaned.
• Command palette (Ctrl+K) — fuzzy-search every open tab in any space, plus your bookmarks and history, with quick actions like new tab, new space, and “ask about all open tabs” (hands every open tab’s title and address to your chat session as context).
• Ctrl+Tab — hold Ctrl and tap Tab to flip through your tabs most-recently-used first, the same feel as Alt+Tab for windows.

Persistent cookies and storage

The browser uses an Electron partition called persist:browser. Everything you’d expect to survive a browser restart in normal Chrome — cookies, localStorage, IndexedDB, cached resources — survives an Omniscio restart here too, scoped to a directory inside your Omniscio user-data folder.

Practical consequence: sign into a site once and you stay signed in. Open Gmail, sign in, close Omniscio, reopen tomorrow → still signed in. This same persist:browser session is also used by the Gmail email-link viewer — the in-app page that opens when you click a link inside an email — so signing into Google in the Browser pane also lets a Google Sheet/Doc linked from an email open signed-in (and vice-versa). It is separate from the cookies the rest of Omniscio stores (the Gmail inbox itself + Google Auth Gate use their own token-based auth / partitions).

The partition is also isolated from your real Chrome profile. Omniscio has no way to import bookmarks, saved passwords, autofill, or sign-in state from external Chrome — you’ll start fresh and have to sign back into each site the first time you visit.

Google sign-in caveat

Google actively detects embedded browsers (Electron webviews, Chromium Embedded Framework, etc.) and blocks sign-in on many flows with a “This browser or app may not be secure” error. Omniscio mitigates this by presenting a stock Chrome identity: a Chrome User-Agent pinned to the browser engine Omniscio actually ships (currently Chrome 148, kept honest by a build guard so it never goes stale), plus the matching Sec-CH-UA Client-Hint headers for Google. Setting the User-Agent string alone isn’t enough — a mismatch between it and the engine-derived Client Hints is exactly what Google’s newer heuristic flags — so Omniscio aligns both to look like a current, consistent Chrome. This is enough for most sites and most Google sub-products after an initial sign-in.

If the sign-in flow itself fails, sign in once in your real Chrome at accounts.google.com, then come back to Omniscio’s Browser pane. For many Google properties the second-factor cookie won’t carry over (different partition), so this isn’t always a fix — as a last resort, use a service-specific token / app password if the site exposes one, or view that property in your real browser.

Opening a Google doc linked from an email? The Gmail email-link viewer shares this same session, so a sign-in done in the Browser pane carries over. If Google still shows its rejection wall for a private doc, the viewer detects it and replaces the dead-end loop with a one-click “Continue in your browser” button that opens the original document link in your real browser — where Google allows the sign-in — so you’re never stuck.

The History button (six tabs in one)

The History toolbar button opens one modal with six tabs:

• History — pages you’ve visited in your own tabs; click one to reopen it, or clear everything (with a confirmation). The address bar also suggests from your history and bookmarks as you type.
• Memories — what Omniscio has indexed for semantic search; see Memories below.
• Passwords — your autofill vault; see Your saved logins below.
• Journeys — your recent browsing auto-grouped by topic; see Journeys below.
• Permissions — per-site rules for what the agent may do; see The safety model below.
• Skills — your saved browser workflows; see Skills below.

How a session drives it

Any Omniscio session automatically gets browser tools — nothing to configure. The toolkit is broad: open a page, read it as clean text, click, type, fill whole forms, scroll, hover, press keys, take or save screenshots, save a page as a PDF, read a YouTube video’s transcript, list every link on a page, watch a dev site’s console and network activity, and open, switch, and close its own tabs. It perceives the page first (a structured snapshot of what’s there), then acts on what it saw — the same look-then-act loop a careful person uses.

The agent’s tabs are marked in the tab strip and live on a completely separate profile. If a task needs the agent logged into a site, it signs in within its own profile — it can never act as you on a site you’re signed into. Its browsing is never recorded to your history, and it can only open normal web pages — it can’t be pointed at files on your computer.

The safety model, in plain words

• Per-site permissions (History → Permissions) — set any site to control (the default: the agent may read and act), read-only (look but never touch), no-access (can’t even read it), or blocked (no-access, meant for sites the agent should never go near, like your bank). A site’s explicit setting always wins.
• Watch-mode for sensitive sites — well-known banks, brokerages, payment processors, and social networks are on a built-in watch list: the agent may act on them only while you’re actually watching (the Browser panel open and focused). If it tries while you’re away, the action is refused and a “the agent needs you watching” card appears in your Inbox — it waits, it doesn’t fail silently.
• Nothing financial happens unattended, ever. Even a guided shopping task stops at the payment step — the agent never enters card details and never places the order. You review and pay by hand.
• Attended by default — beyond the watch list, agent actions currently require you present at all. A scheduled or background run can read pages (pull numbers off a dashboard, say) but can’t click or submit anything.
• Plan approval — before a multi-step task, the agent posts its plan as a short numbered checklist to your Inbox for approval. No answer means it keeps waiting — a plan is never auto-approved.
• Pause and take over — a Pause button in the toolbar freezes all agent actions instantly so you can take the wheel; Resume hands it back.

Your saved logins (autofill vault)

History → Passwords is a small password manager for your own tabs. Log into a site and Omniscio offers to save the login — you type it into the save dialog yourself; it doesn’t capture what you typed on the page. Come back to that site later and a toast offers to fill it in. Passwords are encrypted through your operating system’s own credential store, and the list only ever shows the site, username, and last-used date — never the password itself.

The agent can never read this vault. No tool and no route reaches it — the same wall that keeps your cookies out of the agent’s tabs keeps your passwords out too. Don’t confuse it with Browser Logins (the sidebar entry under Agent Tools), a separate feature that saves a browser session — never a password — specifically so your agents can reuse it. Short version: the autofill vault is for you only; Browser Logins is for the agents.

Memories (ask about pages you’ve already seen)

Omniscio quietly indexes the pages you visit in your own tabs — never the agent’s — entirely on your machine. A session can then search everything you’ve browsed: “what was that pricing page I looked at last week?” comes back with the page, the matching text, and when you saw it, instead of you digging through history by hand. Manage it from History → Memories: browse what’s indexed, delete individual entries, clear everything, or pause indexing entirely (it’s on by default).

Journeys (your browsing, auto-grouped)

History → Journeys groups your recent browsing into topic clusters automatically — nothing to save by hand. Handy for “what was I looking at when I was researching flights?” It’s a light, on-the-fly grouping, not a precise research tracker.

Design mode & Visual Edit (point at the page, change your code)

Two toolbar tools for when the page you’re looking at is your own project, both working with the session bound in the side chat:

• Visual Edit (cursor-target icon) — click any element on the page and describe the change in plain English: “make this button blue.” Omniscio sends your session everything it needs — the element, its styles, a cropped screenshot — and the session makes the change in your source code. No session bound? The same prompt is copied to your clipboard to paste anywhere.
• Design mode (palette icon) — click an element and a live style panel opens: color, background, font size and weight, spacing, size, corner radius, plus move-up/move-down among its neighbors. Changes preview instantly on the real page so you can see exactly what you’re doing; Apply sends just what you changed to your coding session with an instruction to edit the source. It edits your code — it never hacks the live page.

Cursor chat & Compose (AI writes into any text box)

The pencil icon captures whatever text field you’re focused on. Select some text first and it opens in rewrite mode (“make this friendlier”); focus an empty field and it opens in compose mode (“draft a polite reply declining”). Type the instruction and your bound session writes the result straight back into the field. In plain text boxes, Ctrl+Z undoes it exactly like your own typing; in rich editors like Gmail it’s best-effort — treat undo there as a bonus, not a guarantee. Password fields are never captured.

Record a workflow, run it as a Skill

The Record button captures your clicks, typing, and page moves while it’s on. Hit Stop and a review panel lists every value you typed — you choose which ones become fill-in-later blanks (so “search for October invoices” can become “search for anything”), and anything that looks like a password shows up hidden rather than being saved as-is. The result is a Skill. An agent can record its own workflow the same way, with the same review step.

Skills (History → Skills) is your gallery of saved browser workflows — “Summarize this page”, “Pull the action items off this doc” — each with a name, plain-English steps, and an optional emoji. Running one goes through the same run dialog and the same safety gates as any recipe — a skill is a shortcut to a workflow, never a way around the guardrails. For now it’s your own skills only; there’s no community gallery yet.

Side chat

A collapsible chat panel docks on the right of the Browser view with whichever session you pick — the same chat as the Dashboard, just parked next to the page so you’re not switching views to talk to the agent that’s driving. This is the session Visual Edit, Design mode, and Cursor chat all hand off to. Toggle it from the command palette or its own collapse control.

Staying signed in

Your own tabs keep the same persistent profile the Browser pane has always used: sign into a site once and you stay signed in across Omniscio restarts. It’s isolated from your real Chrome — bookmarks, passwords, and sign-in state don’t carry over, so you’ll sign into each site fresh the first time you visit.

One caveat: Google sometimes blocks sign-in inside embedded browsers with a “This browser or app may not be secure” error. Omniscio works around it well enough for most sites; if a Google sign-in still fails, sign into that Google property in your real browser, or use a service-specific app password if the site offers one.

What’s intentionally missing

• No extensions. Ad blockers and third-party password managers don’t apply here.
• The agent stays on the web. It can only open normal http/https pages — it can’t be pointed at files on your computer.
• Design mode reordering is step-by-step (move up / move down among neighbors), not drag-and-drop.
• Cursor chat in rich editors is best-effort — plain text boxes are the fully-supported case.
• Recorded Skills replay literally. A recording repeats what happened, generalized only by the blanks you tagged — it doesn’t infer a broader workflow from one run.

It’s still not meant to replace your daily-driver browser — it’s the browser your AI can safely help you in.

Running Apps is a built-in sidebar virtual hub (emerald app-window icon). When a Claude session spins up a dev server — a frontend, an API — it shows up here so you have one place to see every live app, open it in-pane, and stop it, without hunting for ports.

It’s a thin UI over portless, a small dev proxy. Instead of each dev server grabbing a random port you have to remember, Omniscio launches it behind portless, which gives it a stable, human-readable hostname like https://my-feature.localhost. The .localhost name always resolves to your own machine, so these URLs only ever reach local dev servers.

Requirements

Two prerequisites: portless installed, and system Node 24 or newer (portless requires it). You don’t have to install portless by hand — if it’s missing, the panel shows a one-click Install portless button (see below). If your Node is older than 24, the panel tells you to update Node first and links to the download, since the install would fail otherwise.

What you see

One row per live app: its short slug, a green port chip, the browsable .localhost URL (click to open it inside Omniscio in an embedded view), and a Stop button. A Refresh button re-reads the list, and the panel also live-updates on its own as apps come and go.

When nothing is running, the empty state adapts to what’s on your machine: portless installed → a plain “No apps running”; portless missing, Node 24+ → an Install portless button (one click, with a spinner, then it’s ready — no restart); Node too old → an “update Node first” message with a download link. Installing portless from here uses Omniscio’s built-in tool installer, so it also appears with an Install button in Settings → Connected Tools and the Tools screen.

Enabling it

Running Apps is off by default. Turn it on at Settings → Features → Running Apps. Toggling it off hides the sidebar row and pauses the live poll; toggling back on restores both.

When you have a Claude session work on a Chrome extension project, this feature lets the agent build and test the extension itself — loading it into a real Chrome, clicking the popup, taking screenshots, reading the console, and reloading after code changes. No manual clicking required.

It works by giving the agent a built-in extension-dev workflow that teaches it the load → exercise → screenshot → reload → iterate loop — exactly the kind of hands-on verification a developer would do themselves. Unlike web-app testing (which uses the lightweight Obscura browser), Chrome extensions need a real, pixel-rendering Chrome that can load unpacked extensions, and that’s exactly what this feature provides.

What happens automatically

When all the conditions are met (your project has a Chrome extension manifest.json, the feature is enabled, and agent-browser is installed), Omniscio automatically:

• Injects the extension-dev workflow into every session opened on that project.
• Writes a agent-browser.json file to the project root — { "headed": true, "extensions": ["."] } — which tells the browser tool to launch in headed mode and load the project as an unpacked extension. This file is added to .gitignore so it is never committed.

What it is not

• Not Chrome Web Store publishing — this covers building and testing only. Submitting to the Store is a separate step.
• Chrome only — Firefox and Safari extension formats are not supported.
• Not general browser automation — the feature is specifically for the Chrome-extension development loop, not arbitrary web browsing or scraping.

Setting it up

1. Open Settings (gear icon in the toolbar).
2. Go to the Lab tab.
3. Flip the Chrome extension dev (Autopilot) toggle on.
4. If agent-browser and Chrome-for-Testing are not yet installed, Omniscio shows a “Set it up for me” button. Click it — Omniscio downloads both in one step. You do not need a separately installed Chrome.

The “Build a Chrome Extension” mission

A guided coaching mission takes you from idea to working, tested extension in a single session. It scaffolds the extension folder structure and manifest, then hands off to a session that has the Autopilot workflow already active — so the agent builds the popup and any background logic, then verifies each piece by driving Chrome itself. Reach the mission from the project empty-state or the mission picker.

Privacy

agent-browser launches a clean managed Chrome-for-Testing instance — not your real Chrome profile. Omniscio passes no cookies, saved passwords, bookmarks, or sign-in state to the test browser. It is physically isolated from your real Chrome.

Extensions hub

When the feature is enabled, an Extensions tool appears in the sidebar. Clicking it opens the Chrome Extensions Hub — a panel listing all your Chrome-extension projects in one place. Each project shows a status pill (Tested if the agent has already run and saved screenshots, Draft if not) and three quick actions: Open (switch to the project), Test it (confirm, then launch an Autopilot session), and Install in Chrome (reveals the folder and walks you through Chrome’s Load-unpacked steps — only enabled after at least one successful test run). Right-clicking a row lets you Rename or Remove the project; Remove is a soft-delete with a one-click undo. The + New Extension button at the top launches the guided “Build a Chrome Extension” mission. Omniscio never auto-installs an extension into Chrome or opens a chrome:// URL — Chrome requires you to load unpacked extensions yourself, so the hub shows you exactly how.

Tasks is Omniscio's built-in task manager. It combines a keyboard-first outliner with AI planning intelligence: a daily check-in that learns your capacity, automatic follow-up extraction from your AI sessions, and coaching for overwhelm and procrastination. Everything is local-only — your tasks, plans, and AI conversations stay on this machine.

Opening Tasks

Tasks is off by default. Enable it at Settings → Features → Tasks, then click the Tasks entry in the sidebar. Turning the feature off later only hides it — your data stays in the database and reappears when you flip it back on.

Core Concepts

• Today plate — the short list of tasks you plan to work on today, separate from the full backlog
• Hubs — group tasks under a hub with master context that cascades to every child task
• Lists — organize tasks into custom lists (Inbox is the default)
• Importance & urgency — rate each task on a 0–3 scale; the AI uses these to prioritize
• Done / Dropped archive — completed and dropped tasks move to the archive, keeping the active list clean

AI Planning Intelligence

Tasks includes several AI-powered planning features, each running as a bounded Claude session:

• Daily check-in — a conversational morning ritual where the AI reviews your plate, suggests a focus order, and proposes new tasks or splits. One check-in per day; the conversation is saved as a planning journal.
• Plan my day — a deeper planning session that produces a structured plan (focus items, new tasks, splits, profile updates) you can accept or modify
• Right Now — a single-item answer to "what should I do next?", combining local ranking with a cached AI verdict
• Arrange — AI-powered reordering of your Today plate with per-pick rationale
• Plan chat — a free-form ongoing conversation about your day's plan, with full context from your plate, backlog, and learnings

Adaptive Memory

Tasks learns from your behavior over time:

• Learnings — short behavioral insights the AI distills from your planning patterns (e.g., "Plans with more than 4 deep-work items historically fail"). You can edit or dismiss any learning; editing makes it permanent so the AI cannot overwrite it.
• Day log — a daily record of what was planned vs. completed vs. carried over, with counts. The AI uses 14 days of this history to calibrate realistic planning recommendations.
• Reality snapshot — a behavioral profile (median planned, median completed, plan rate, chronic carryovers, catch keep rate) that primes every AI prompt so recommendations reflect your actual capacity.

Caught Items

When an AI session ends, Omniscio scans the transcript and automatically extracts follow-up tasks, detects completed open tasks, and identifies stale suggestions. Caught items appear in a review queue — keep, edit, or dismiss each one. The system calibrates over time based on your keep rate.

Suggestions older than 30 days are automatically hidden from the active list so it stays focused on recent items. They are not deleted — a banner shows how many were hidden and links to Ask Omniscio, where you can search for them and re-add any you still need.

Coaching Sessions

Three coaching personas are available as dedicated AI sessions:

• Overwhelm Coach — a two-act session: emotional regulation first (breathing, brain dump, ruthless triage), then structured planning
• Breakdown / Momentum — breaks a large task into phased steps with calibration
• EAD Analyzer — Eliminate, Automate, Delegate analysis for your task list

The Outliner

• Each row is one task with a checkbox, the task text, and optional chips for due date, snooze, importance, and urgency
• Children indent under their parent to any depth
• A breadcrumb across the top shows where you are when you drill into a sub-tree
• Repeating tasks spawn a successor when completed or archived

Keyboard Shortcuts

Agent and CLI Access

Tasks exposes a full CLI API (85+ endpoints) for agents and automation. Agents can create, update, complete, and organize tasks; read the Today plate; trigger AI planning sessions; and poll for fired events. All CLI access requires the local bearer token and is rate-limited.

A read-only Markdown mirror (tasks-v2.md) is written to your user-data folder on every save, so local agents can read your task list as a plain file.

Notifications

When a task's due date or snooze time arrives, Omniscio fires a notification. Focus Mode batches reminders into a single grouped toast when active.

Scratchpads is a tiny "quick notes" surface that lives as a virtual hub in the Omniscio sidebar. Two panes: pad list on the left (most-recently-updated first), rich body editor on the right. No folders — it is deliberately the smallest possible "jot it down" tool, not a notebook — but you can paste images inline alongside your text, and a header toggle adds live markdown styling (headings, bold, italic, code, lists, blockquotes, links) while still letting you edit. Local-only: every pad lives in Omniscio's database on this machine, and pasted images are embedded directly in the pad. Nothing leaves your computer.

Opening Scratchpads

Click the Scratchpads entry in the Omniscio virtual-project group in the sidebar. The first time you open it, the panel shows a "No scratchpads yet" card with a Create button.

The Two-Pane View

• Pad list (left) — one row per pad, sorted with the most recently edited at the top. A filter box at the top of the list narrows by title and body text, case-insensitive.
• Editor (right) — the header bar holds an inline-editable title (defaults to the pad's first line), a Saved pill, a Markdown toggle button (paragraph-icon, see below), a Copy all button, and a Delete button. The body accepts plain text and pasted images.
• Stats row — a thin footer under the editor shows chars · lines · ~tokens (and N images when at least one image is pasted in). Tokens are an estimate (chars ÷ 4, marked with ~); image data is excluded from the char count so a screenshot doesn't inflate it.
• Auto-save — 500 ms after you stop typing, Omniscio writes the pad to the database. The pill flips to "Saving…" while in flight and back to "Saved" when done.

On a phone (or any narrow window) the two panes stack into one: you see the full-width list of notes, and tapping a note (or +) opens it full-screen with a ← Scratchpads button to return to the list. On a desktop-width window the two panes sit side by side as described above.

Pop Out Into Its Own Window

Click the Pop out button (the small open-in-new-window icon — in the two-pane view's header next to +, and in the Quick Capture overlay's header) to move your scratch pad into its own separate, movable, resizable window. You can also open it by right-clicking the Scratchpads row in the hubs sidebar and picking Open in new window. Park it on a second monitor or beside whatever else you're doing — it stays open while you use the rest of Omniscio.

• Same notes, live — the window shows the full list-and-editor view, and edits sync both ways with the in-app scratch pad instantly (it's the same notes underneath).
• One window — clicking Pop out again just brings the existing window to the front rather than opening another. Close it like any normal window.
• Remembers itself — it reopens at the same size and position, and follows your light/dark theme automatically.
• Popping out from the Quick Capture overlay saves whatever you've jotted first, then opens the window on that note.

(Desktop only — the button isn't shown on a phone.)

Pasting, Copying, and Resizing Images

Take a screenshot (Windows: Win+Shift+S, then Ctrl+V into the pad) and the image lands inline at the cursor. To get an image back out or change its size:

• Single-click an image to surface two small overlays: a Copy image chip at the top-right (click it to copy to the system clipboard) and a tiny resize handle at the bottom-right corner.
• Drag the resize handle (the cursor changes to a diagonal arrow) to scale the image. The aspect ratio is preserved automatically — you only drag horizontally. There is no max height; you can grow the image up to the editor column's width.
• Right-click an image for a small context menu with Copy Image and Remove.
• Scrolling or typing dismisses both the Copy chip and the resize handle.
• Ctrl+Z reverts a resize the same way it reverts a text edit — the resize is captured in the pad's undo history.

The Copy chip and resize handle both use theme-aware colors, so they stay visible whether you're in light or dark mode.

Rich-text paste from a webpage is reduced to plain text — scratchpads are not a rich-text editor.

Markdown Styling

Click the small paragraph-icon button in the editor header to turn live markdown styling on or off. When it's on (the button is filled with the accent color), Omniscio visually styles your text as you type:

• Headings — lines starting with #, ##, or ### render bigger and bolder.
• Bold — text wrapped in **double asterisks**.
• Italic — text wrapped in *single asterisks* or _underscores_.
• Inline code — text wrapped in `backticks`.
• Code blocks — lines starting with three backticks (```) open and close a code block.
• Lists — lines starting with - or *.
• Blockquotes — lines starting with >.
• Links — [link text](https://example.com) displays the text styled as a link.

You're still editing the underlying markdown source — turning the toggle off doesn't remove the ** or # characters, it just stops styling them. The toggle preference is saved across app restarts, and the Quick Capture overlay (Ctrl+Shift+S) has the same toggle button in its header — both flip the same setting, so turning it on in one place turns it on everywhere.

List shortcuts (bullets & numbering)

Two shortcuts turn the line your cursor is on into a list — the same keys the Vault editor uses. They work everywhere you edit a scratchpad (the main view, Quick Capture, and the popped-out window), whether or not markdown styling is turned on.

• Bullet list — Ctrl+. (or Ctrl+Shift+8, the Word / Google‑Docs combo).
• Numbered list — Ctrl+/ (or Ctrl+Shift+7).

Press once to add the marker, press the same keys again to remove it. Select several lines first and numbering counts them 1, 2, 3 down the list. Press it on a heading and it cleanly becomes a list instead of stacking. Your place in the text is kept as the marker appears, and it's a single Ctrl+Z to undo.

Quick Capture (Ctrl+Shift+S)

Press Ctrl+Shift+S from anywhere in Omniscio — even while typing in another input — to open a large centered overlay — roughly 80% of the viewport width and 85% of its height, a comfortable near-full-screen note surface, but still a modal with a visible backdrop — for a quick note. The Scratchpad button (NotebookPen icon) is also pinned to the top-right toolbar by default, so you can reach the overlay with a click. The footer hint "Press Ctrl+Enter or close to save. Notes shorter than 3 characters are discarded." is centered under the editor, with the same chars / lines / ~tokens / images stats row above it. Esc, click-outside, the X button, Ctrl+W / Cmd+W, or Ctrl+Enter closes it (Ctrl+W is intercepted by the overlay so it doesn't archive the session behind it).

The overlay opens as a single editor surface — the editor fills the modal, both its full width and its full height (from the toolbar down to the stats row). A pad list is available behind a header toggle so you can swap between recent pads, but it stays out of the way by default:

• The list is collapsed by default, every single time you open the overlay (no persistence). To reveal it, click the "N notes" pill — a pill-shaped toggle in the second row of the header (beneath the title) showing a panel icon and a live count, e.g. 37 notes. Clicking expands the pane, flips the icon, and tints the pill in the accent color; click again to hide. The count lives right in the pill text (no separate badge, no chevron), so it always tells you how many notes you can open. You can also toggle the list with Ctrl+L, or open it with Tab and close it with Shift+Tab. Toggling the list doesn't remount the editor or drop typed content.
• When expanded, the list pane shows every pad sorted most-recently-updated first, with title, relative time, and a snippet of the body. Click any row to swap the editor to that pad in place — the overlay stays open and the list stays open.
• The + New button at the top of the list pane clears the editor to fresh-draft mode (no pad selected). The next save creates a new pad, and clicking + New also clears the 60-second reattach window so a quick "wait, one more pad" never collapses two pads into one.

On close:

• In fresh-draft mode (the default each time you open the overlay), if the visible plain text has fewer than 3 characters (after trimming), nothing is saved — the overlay just closes. A pad consisting only of a pasted image is kept (the image is the visible content). Otherwise, a new pad is created with the first line as the title.
• In existing-pad mode (after clicking a row in the list pane), close flushes any pending autosave against that pad and does NOT create a new one.
• If you re-open the overlay within 60 seconds of a fresh-draft close, Omniscio re-attaches to the last pad instead of creating a new one — so a quick "wait, one more line" doesn't fragment a note across two pads. Switching pads inside the overlay or clicking + New clears this window so intra-overlay flows can never accidentally reattach.

The overlay header is two rows. The top row has the "Quick scratchpad" title on the left and the X close button on the right. The second row has the "N notes" pill (show / hide the pad list, described above) on the left, and on the right a Markdown toggle (the same paragraph-icon button described above) followed by a Copy all button. The markdown toggle flips the same global preference used by the main editor, so live markdown styling lights up immediately in the overlay as you type. The Copy button copies the current overlay body to your OS clipboard as plain text (HTML wrappers from the rich editor are stripped) and is disabled while the body is empty and image-free, so an accidental click on an empty overlay never overwrites the clipboard.

Overlay shortcuts toggle the layout without touching your text: Ctrl+L shows or hides the pad list (same as the "N notes" pill), and Tab toggles the list (press once to show it, again to hide it) while Shift+Tab always closes it. They work even while you're typing in the note, and the setting doesn't stick — the list starts collapsed every time you open it. (Inside this overlay, Tab is dedicated to the pad list, so it no longer hops between the overlay's buttons.)

The keyboard shortcut is rebindable at Settings → Keyboard Shortcuts → Quick scratchpad.

Save and start new (Ctrl+T inside the overlay)

Press Ctrl+T while the Quick Capture overlay is open to save the current note and reset the editor to a blank fresh draft — without closing the overlay. Use this when you want to jot a series of separate notes back-to-back, one after another, without each one closing and re-opening the surface.

• In fresh-draft mode, if the body has at least 3 visible characters (or any pasted image), a new pad is created. Sub-threshold bodies are discarded, the editor still clears. Same "not worth keeping" rule the close-on-Esc path uses.
• In existing-pad mode (after picking a row from the list pane), any pending autosave is flushed against that pad and the editor swaps back to a fresh draft. The just-edited pad stays in the list with your edits saved.
• After every successful Ctrl+T, the overlay stays open, the editor remounts empty, and the 60-second reattach window from the close-and-reopen flow is cleared — so the next save creates a brand-new pad, never silently overwriting the one you just saved.
• If the save fails (e.g. database locked), an error toast appears and your draft is preserved so you can press Ctrl+T again without losing the text.
• Two rapid Ctrl+T presses coalesce silently into one save — the second press is ignored while the first is in flight.

Outside the overlay, Ctrl+T keeps its normal meaning — launching a new Claude session in the active project. The two interpretations never fire together: whenever the overlay (or any other modal) is open, the global new-session shortcut is suppressed.

Delete, Trash & Restore

The trash icon in the editor header opens a confirm dialog. After deleting, the pad moves to Trash and a toast appears with an Undo button — click it to restore the pad immediately.

Trash. The header bar shows a Trash button with a count badge (e.g. “3”) whenever trashed pads exist. Click it to open a popover listing your trashed notes, newest-deleted first. From there you can Restore a pad (moves it back to your active list) or Delete forever (permanent, no recovery).

30-day auto-purge. Trashed pads older than 30 days are automatically and permanently deleted. This is always on — no toggle. The 30-day window gives you time to recover an accidental delete.

Auto-delete (optional). Under Settings → System → Auto-Delete Old Scratchpad Notes you can have notes you haven’t edited in a chosen number of days (1–365, default 30) moved to Trash automatically. It is off by default and asks for confirmation before you turn it on. Auto-deleted notes land in Trash and can be restored within 30 days (after which the auto-purge removes them). Editing a note resets its timer.

Still deferred

The following are deferred to a later release: pinning, right-click "Save message to scratchpad" from a chat, global Ctrl+K search across pads, and a settings toggle to disable the feature.

Writer Studio is a standalone writing surface inside Omniscio. It opens as a virtual hub in the sidebar and gives you a clean, full-screen editor for writing documents in markdown — notes, drafts, documentation, anything. Documents are saved locally on your machine (they are not tied to any Claude account) and auto-save as you type.

Turning it on

1. Open Settings (gear icon in the toolbar).
2. Go to the Lab tab.
3. Flip the Writer Studio toggle on.
4. A Writer Studio entry appears in the Omniscio virtual-project group in the sidebar. Click it to open the writing surface.

Creating and opening documents

• The document list is on the left. Click + New Document to create one.
• Each document has a title (editable inline) and a body that you edit in the main panel on the right.
• Click any document in the list to open it. The editor loads its content immediately.

Auto-save

Your work is saved automatically — 500 ms after you stop typing, Omniscio writes the document to its local database. You do not need to press Save. Switching to a different document flushes any pending save first, so nothing is lost on quick switches.

Renaming a document

Click the document title at the top of the editor to edit it, or click the … (three-dot) menu on a document in the sidebar list and choose Edit title. Either way, press Enter or click elsewhere to save the new name; press Escape to cancel.

Writing Guidance panel

The Writing Guidance panel lets you write instructions for the AI — things like "keep the tone formal" or "this is for a technical audience." These instructions will be used when AI editing tools are available (coming in a later release). You can fill this in now so it is ready when AI features arrive.

Creativity Dial

The Creativity Dial controls how conservative or inventive AI suggestions will be, on a scale from 0 to 100. Lower values produce suggestions that stick close to your existing text; higher values allow bolder rewrites. Like the Guidance panel, this setting is ready now and will take effect when AI editing features ship.

Deleting a document

Open the document you want to remove, then click the Delete button (trash icon). A confirmation dialog appears — confirm to delete. A toast appears for a few seconds with an Undo button. Click Undo to restore the document before the toast disappears. After that, the document is gone.

What's coming next

The current release covers the core document management loop. AI-powered features are planned for later phases:

• Selection edits — select a passage and ask the AI to rewrite, expand, or condense it.
• Options — generate multiple alternative versions of a section.
• Autocomplete — AI continues your sentence as you type.
• Chat assistant — a side panel for free-form conversation about the document.

Screen Recordings is the built-in screen recorder inside Omniscio. Capture a screen or a single window, review the result in a built-in player, and publish a shareable link — without leaving Omniscio or installing a separate tool. It lives as its own entry in the Omniscio sidebar (a film-frame icon, in the Omniscio built-ins group). Everything is local-first: recordings are written to your own machine and nothing is uploaded unless you explicitly publish.

Opening Screen Recordings

Click the Screen Recordings entry in the Omniscio sidebar. The library view opens — a list of every recording you've made, newest first. The first time, the list is empty and shows a “No recordings yet” message. If you don't see the sidebar entry, enable it at Settings → Features → Screen Recorder (on by default).

Starting a Recording

Open the Record tab (or click New recording in the library header). Starting a recording is a two-step flow, with a live preview of your composed recording always visible at the top — it updates the moment you pick or arrange anything, on both steps:

• Step 1 — Choose what to record. A grid of tiles, one per screen and per open window the OS exposes, a Record my webcam toggle with a camera picker, and your saved presets & templates (loading a preset jumps straight to Step 2 with the saved scene). Select one or more sources; the Next button stays disabled until you do.
• Step 2 — Recording options. Every option: output size, resize mode, layout arrangement, microphone and system audio, and webcam shape. Back returns to Step 1 at any time — your picked sources and arranged options are kept.
• Start recording at the bottom of Step 2 begins the capture with your composed scene.

While the picker looks for available sources it briefly shows “Finding screens and windows…”. This step is bounded to about 8 seconds. On some machines — remote desktop (RDP), a locked screen, or a graphics-starved environment — the operating system's source list can otherwise hang forever, leaving you stuck on the spinner. Omniscio now always settles within ~8 seconds: if no sources came back it shows a “No capture sources available” message instead of hanging. If you hit that, check that a display is connected and that Omniscio has the OS screen-recording permission, then try again. On a normal desktop the list appears almost instantly.

While Recording

Once capture starts it keeps running in the background no matter which Omniscio view you're looking at. A floating control bar appears on the same display as the Omniscio window — and you can drag it anywhere (grab the bar anywhere that isn't a button) if it's covering something you want to record around. The controls:

Review & Publish

Click any finished recording in the library to open the Review Window:

• An inline video player that plays the local MP4
• A title field (defaults to the source label, e.g. “Screen 1” or a window title)
• An optional custom URL slug (lowercase letters, numbers, and hyphens; up to 64 characters)
• Publish — marks the recording published and gives you a shareable link to copy (Ctrl+Enter also submits). After publishing, the footer swaps to Copy link + Done.
• Discard — permanently deletes the recording (after a confirm dialog)

Recovered Recordings

If Omniscio closes or crashes mid-recording, the partial capture isn't lost. On the next launch, Omniscio scans for interrupted recordings and surfaces them in the library as “Recovered” entries — click one to finish processing it into a normal MP4, or discard it.

Where Recordings Live

Every recording is stored under a screen-recordings folder in Omniscio's user-data directory on this machine:

MemPalace (Memory Palace) gives your AI agents persistent memory across sessions. Normally, each AI session starts fresh with no memory of previous sessions. MemPalace changes that by letting agents store and recall information.

How It Works

When MemPalace is enabled, Omniscio injects a memory server (via MCP) into each session. The AI agent can:

• Store memories — Save important information, decisions, patterns, or preferences
• Recall memories — Search for relevant information from past sessions
• Update memories — Correct or refine stored information

Setting Up

1. Go to Settings → Features.
2. Toggle "Enable Memory Palace" on.
3. New sessions will automatically have access to the memory system.

Viewing Memories

Once enabled, open the toolbar’s “more” (…) menu and click Memory Palace (under Notes & capture) — or pin it to the toolbar — to browse stored memories. You can search, view, edit, and delete individual memories.

The Vault is an integration that lets you point Omniscio at a folder of Markdown notes (a "vault") and use it like a built-in editor. You browse notes in a file tree, open them in a TipTap-based rich editor, drop images straight into the body, search across every note's title + content + images in one place, and let Omniscio's AI describe / OCR / tag your images automatically. It is gated by a setting, off by default, and writes only inside the vault folder you pick.

Quick Reference Wizard

The first time you enable The Vault, a Quick Reference Wizard opens automatically to walk you through the basics: what The Vault is, how notes and search work, what agent tools are available, how to ask your vault questions, and power features like the pop-out window and AI summaries. The final step lets you pick your vault folder right inside the wizard.

The wizard is purely educational; closing it early is fine. You can replay it any time from two places:

• Settings → Setup Wizards — a "Vault Quick Reference" card with a Take/Retake button (shows a green checkmark if you have completed it before).
• Settings → Features → The Vault — a "Quick Reference Guide" card near the top.

If you already had The Vault enabled before this feature was added, you will see a one-time inbox alert letting you know the guide is available.

Turning The Vault on

1. Open Settings → Features and flip Enable The Vault.
2. In the Vault card on that same Settings → Features page, click Choose folder… and pick the folder that holds your .md notes (e.g. C:/MyNotes). Omniscio checks the folder is usable first — it must already exist and be a normal folder, not a system folder or a shortcut — and tells you plainly if it isn’t, so nothing is saved until you pick a good one. Changing the folder takes effect right away — the Vault panel switches to the new folder’s notes, no restart needed. (You can also turn Watch vault for changes off here if you’d rather Omniscio only re-scan on restart.)
3. Omniscio's indexer scans every `.md` file under the vault root and parses YAML frontmatter, #hashtags, and [[wiki-links]]. The first scan takes a moment; subsequent launches are incremental.
4. A new The Vault entry appears in the Omniscio sidebar group with a book icon. Click it to open the panel.

The Vault panel

The panel has a sidebar, an editor, and an on-demand side drawer:

• File tree (Omniscio sidebar, Pane 2) — lives in Omniscio's native sub-sidebar (same slot as Tools, Skills, Alarms, AI Coaching), so it gets the standard resize handle and mobile drill-down. The sidebar has a Files / Sessions toggle at the top; the Files tab shows every folder and `.md` file under the vault root, each note labelled by its name with the `.md` extension hidden (it's implied — every note is a Markdown file). Click a note to open it in a tab. Right-click a note for Rename / Move / Hide / Delete. Right-click a folder for New note / New folder. Add a note with the header's + button (or a folder's New note) — the new note opens, and the tree scrolls to it and highlights it so you can see where it landed, even when your sort order would otherwise place it off-screen.
• Sessions (Omniscio sidebar, Pane 2 — the Sessions tab) — switch the toggle to Sessions to list the AI sessions running in this vault, and start a new one with + New. Each session row behaves exactly like a session in Omniscio's main sidebar: middle-click to archive it, right-click for the full menu (archive, snooze, pause, move), a hover ··· actions button, a coloured status dot, and Ctrl+Z (or the toast's Undo) to undo an archive. Available on the phone layout too — see On a phone or narrow window below.
• Editor (main panel) — a tabs strip on top (one tab per open note, persisted across relaunches), the TipTap rich editor in the middle, and a status bar at the bottom showing word count + character count.
• Side panel drawer (closed by default) — click the Toggle side panel button in the editor toolbar's View group (the rightmost icon, shows a panel-on-the-right glyph) to slide out a stack of collapsible panels: Outline, Backlinks (every note that wiki-links to the active note), Smart Connections (notes semantically related to the one you're reading but not yet linked — requires Semantic Search; see below), Tags, Bookmarks, Hidden (notes you've hidden from the tree), and Gallery (every image in your vault — see the Images section below). The drawer slides in smoothly from the right, over the editor only — the tabs and toolbar stay put above it, so the toggle (and your formatting buttons) stay visible and usable while it's open. Click the X in the drawer header (or the Toggle side panel button again) to slide it away without leaving the editor; your last open/closed choice is remembered across relaunches.

On a phone or narrow window

On a small touch screen the panel switches to a layout built for one-handed use. A compact top bar shows ☰ Notes, the current note's title, and a ⋯ More button. Tap ☰ to slide your note list (the file tree) in over the editor; tap a note and it opens and the list slides away. The open-note tabs become a single row you swipe sideways. Tap ⋯ (or Aa More on the slim bar that appears while you type) to open a sheet holding every formatting control — bold, italic, headings, lists, links, images, and the rest — plus the side-panel, find, source-view, and bookmark actions. The desktop toolbar and status bar are tucked away to keep the writing surface calm. The sidebar's Files / Sessions toggle is here too — tap Sessions to reach the AI sessions running in this vault and tap one to open its chat full-screen, the same sessions you'd see on the desktop.

Vault Overview — every note at a glance

The Vault Overview is a bird’s-eye table of every note in your vault on one screen — title, tags, its AI summary, an estimated size (tokens and bytes), and when it was last edited. It’s the companion to the note-by-note editor: handy for auditing a big vault, finding a note by tag or summary text, and seeing at a glance which notes are large.

• Open it — the Vault Overview button in the editor toolbar, or Ctrl+Shift+P → “Vault Overview”. It also has its own rebindable shortcut at Settings → Keyboard Shortcuts.
• Search — the filter box narrows the list by title, tags, or summary text as you type.
• Sort — click a column header to sort by title, size, or last-edited; click again to reverse.
• Open a note — click any row to open it in the editor, just like clicking it in the file tree.
• On a phone — the table becomes a stacked card list so it stays readable on a narrow screen.

Give the Vault its own theme

The Vault can wear a different look from the rest of Omniscio — handy if you want your notes to feel like a calm reading room while the app keeps your usual theme. Open Settings → Features → The Vault and find the Vault appearance block. Everything there defaults to follow the app, so nothing changes until you pick something.

• Visual theme — choose any of Omniscio's built-in themes or one of your own imported custom themes for the Vault only. The editor, its file tree, and the pop-out window all adopt it; the rest of Omniscio stays on your global theme.
• Light / dark — give the Vault its own light or dark mode (or leave it following the app), so you can keep a light reading vault inside a dark app.
• Editor font — Default, Serif, or Monospace for the note text.
• Editor text size — Small, Medium (default), Large, or Extra Large. Independent of the app-wide font size, so you can keep tiny UI chrome but large reading text, or vice versa.
• Editor width — Narrow, Normal, Wide, or Full. Full (the default) keeps today's edge-to-edge editor; the others center a comfortable reading column.
• Page background — Follow theme, Paper, or Sepia. Paper and Sepia give the writing area a warm, easy-on-the-eyes page even when your theme is dark.
• Cross off completed checklist items — when on (the default), checking a task-list checkbox dims the text and draws a line through it. Unchecking restores the original look.

You can also change the text size and light/dark mode without leaving the editor: click the gear icon in the editor toolbar (between Ask AI and Bookmark) to open a quick-access appearance popover. Changes apply immediately and persist across sessions.

Editing a note

• Open — single-click any note in the file tree.
• Type — the editor is a TipTap-based rich Markdown surface. Bold / italic / headings / lists / code blocks / tables / blockquotes / highlight all work via toolbar buttons or markdown shortcuts (e.g. `# ` for H1, `- ` for bullets, `==text==` for highlight).
• Autosave — 500 ms after you stop typing, Omniscio writes the note back to disk. It runs quietly in the background — there's no save/“Unsaved” status indicator to watch, and only a failed write raises a notification. If Omniscio crashes before the save completes, a heartbeat draft is recovered on the next launch (see Privacy & safety below).
• Find / Replace — Ctrl+F opens the find bar, Ctrl+H opens it with replace. The bar has three tabs — This Note, All Documents, and Images — over one search box; press Tab to cycle them, and each shows a live result count. This Note searches the open note in both the rich and source/Markdown views, with a match counter, Enter / Shift+Enter to step through hits, case / whole-word / regex toggles, and replace / replace-all. All Documents and Images search your whole vault — Enter on a result opens it (a note opens with your term still highlighted; an image opens in the lightbox). Esc closes.
• Source view — Shift+Ctrl+M toggles between the rich editor and a raw Markdown textarea (useful for hand-editing YAML frontmatter or pasting in pre-formatted content).
• Tables — insert a table from the ⋯ More toolbar menu → Insert table, or type /table in the editor. Tables use standard GitHub-Flavored Markdown pipe syntax with column alignment (left / center / right). Press Tab to jump to the next cell and Shift+Tab for the previous cell. Add or remove rows and columns from the table controls that appear on hover.

Footnotes

The editor supports standard Markdown footnotes — the same [^1] syntax used by GitHub, Obsidian, and other Markdown tools. Type [^label] inline to create a reference (it renders as a clickable superscript), and type [^label]: text at the start of a line to create the matching definition. Click a reference to jump to its definition; click the definition’s label to jump back.

• Multi-line definitions — indent continuation lines with 4 spaces or a tab. Blank lines within are preserved for multi-paragraph footnotes.
• Labels — any combination of letters, numbers, hyphens, and underscores (e.g. [^1], [^my-note], [^ref_2]).
• Editing — definitions appear as styled read-only blocks in the rich editor. To edit a definition’s text, switch to source view (Shift+Ctrl+M) and edit the raw Markdown directly.
• Round-trip safe — footnote syntax survives save and reload without any changes.

Callout blocks

Callouts are styled admonition boxes — note, tip, warning, danger, and more — that use the same syntax as Obsidian, so notes with callouts transfer freely between Omniscio and Obsidian.

Insert a callout in three ways:

• Toolbar — click ⋯ More formatting in the editor toolbar and pick Insert callout.
• Slash command — type / and pick any type (e.g. Warning callout, Tip callout).
• Command palette — Ctrl+Shift+P → search “callout” → pick a type.

Callouts support 12 types, each with a distinct color and icon: note (blue), info (blue), tip (teal), warning (amber), danger (red), todo (blue), bug (red), example (violet), quote (gray), success (green), question (amber), abstract (cyan). Obsidian alias names (like hint, caution, error) are also recognized.

Foldable callouts — in source view, add - after the type to start collapsed, or + to start open but foldable:

> [!tip]- Click to expand
> Hidden content here.

The content area inside a callout supports all the same formatting as the rest of the editor — bold, italic, links, lists, code blocks, and more.

Wiki-links and the [[ popover

Type [[ anywhere in a note and a small popover appears below the caret listing matching note titles. Keep typing to filter; press Up / Down to move the highlight, Enter or Tab to pick the highlighted row, Esc to dismiss. The picker inserts [[Title]] at the caret. If no existing note matches your typed text, the popover offers a Create new: "<Query>" row at the bottom — picking it creates the note for you in one keystroke and inserts the wiki-link into the source note.

When you rename a note (right-click → Rename), Omniscio rewrites every [[wiki-link]] reference to that note across the whole vault in a single pass — links never dangle.

Hide vs Delete

• Hide — right-click a note → Hide. The file stays on disk untouched, but disappears from the file tree, search, and link suggestions. Hidden notes are reachable through the Hidden panel in the right rail. Unhide from there to bring them back.
• Delete — right-click a note → Delete. The file is moved to <vaultRoot>/.trash/<timestamp>__<basename> rather than unlinked. Recoverable by hand if you change your mind.

Smart Connections

The Smart Connections panel in the right-pane drawer shows notes that are semantically related to the one you’re reading but not yet linked. It uses note embeddings from Semantic Search to rank results by similarity.

• Up to 20 related notes ranked by similarity, each showing the title, folder path, and a body preview snippet.
• Notes you’ve already linked via [[wiki-link]] and hidden notes are excluded — you only see connections you haven’t made yet.
• Hover a row to reveal a Link button that inserts [[Title]] at your cursor position in one click.
• Requires Semantic Search to be enabled (the embeddings it relies on come from that feature). Without it, the panel shows a helpful message instead of an error.

Images (paste, drop, picker)

Three ways to put an image into a note:

• Paste — Ctrl+V while the editor is focused. Works for screenshots, copied-from-browser images, anything in the clipboard.
• Drag & drop — drag image files from File Explorer or another window straight into the editor.
• Picker — click the image button in the editor toolbar and pick a file from disk.

However the image arrives, Omniscio hashes its bytes (SHA-256), stores it under <vaultRoot>/assets/<hash>.<ext>, and references it from the note. The same image is stored only once — pasting the same screenshot into ten different notes leaves you with one file on disk and ten lightweight references. Pasting the same screenshot a second time reuses the existing file.

Gallery panel

The right rail carries a Gallery panel that lists every image in your vault as a thumbnail grid, sorted most-recently-added first. Click a thumbnail to open the full-resolution Lightbox. The Gallery also gives you:

• A badge on each thumbnail showing whether the image has been analyzed by Omniscio's image analyzer (see below).
• An Analyze all button at the top to kick off bulk AI analysis on every un-analyzed image in one pass.
• A Clean up orphans affordance for images on disk that no note references — one-click move to .trash/assets/.
• Per-row hover: a delete button. If the image is still referenced by any note, Omniscio refuses and lists the referencing notes; once unreferenced, the file moves to .trash/assets/<timestamp>__<hash>.<ext>.

Lightbox

Open the Lightbox by clicking any thumbnail in the Gallery, or by clicking an image hit in the Unified Search modal. Inside:

• Full-resolution image, fit-to-viewport.
• Esc closes; Left / Right Arrow step between images; Z toggles native-pixel-size view.
• Side rail showing Omniscio's AI analysis (if present): a short title, a 1–3 sentence description, every legible character of OCR'd text, a strip of single-word tags, and a high-level category.
• Footer: the image's SHA-256 hash (with a Copy button), pixel dimensions, file size, modification date, and on-disk path.

AI image analysis

Omniscio can describe your images for you. Trigger it per-image from the Lightbox, or in bulk from the Gallery's Analyze all button. For each image, the analyzer asks Claude (via your configured API-key account) to extract:

• Title — a 3–7 word noun-phrase summary.
• Description — 1–3 sentences of prose describing what's visible.
• OCR text — every legible character the model can extract (great for screenshots of dashboards, error messages, hand-written notes, slide decks).
• Tags — single-word tags (e.g. ["screenshot", "diagram", "ui"]).
• Category — high-level bucket (e.g. "screenshot", "photo", "diagram").

Results are stored once per image. Re-analyzing the same image overwrites in place. The per-call spend lands in Omniscio's cost dashboard under the kms-image-analyzer source. Bulk-analyze skips already-analyzed images automatically, so an interrupted run can resume without re-billing.

Ask AI about your notes

With a note open, the editor's toolbar shows an Ask AI button (a ✨ sparkle). Click it for three one-click actions:

• Summarize this note — get a quick summary of the note you're looking at.
• Find related notes — Omniscio searches your vault for other notes on the same topic.
• Ask a question — ask anything about your whole vault.

Whichever you pick, Omniscio starts an AI chat (it opens in the Vault Sessions view) that can read and cite your actual notes to answer — so the reply points you at real pages in your vault rather than guessing. The same three actions are also in the command palette. You need a vault set for this to work; with no vault, the button is greyed out.

Agent tools — let your AI sessions read and write your notes

The Vault can give every Claude Code session Omniscio spawns direct access to your vault through MCP tools. There are two tiers, each behind its own toggle:

• Read tools (Settings → Features → "Expose Vault to spawned agents") — five tools that let agents search, read, list tags, list recent notes, and follow backlinks. Read-only, zero API cost (they hit local SQLite, not the Anthropic API).
• Write tools (same panel, "Allow agents to write to Vault" — only visible once read tools are on) — four tools that let agents create new notes, append content, update frontmatter properties, and insert content at specific positions.

Safety controls for write tools

Write tools have three layers of protection against runaway agents:

• Rate limiting — each session is limited to 10 writes per minute and 200 writes per session (both configurable). When the limit is hit, the tool tells the agent to slow down and how long to wait.
• Folder allowlist — you can restrict which vault folders agents are allowed to write into. Leave it empty to allow all folders, or list specific folders (e.g. drafts/, agent-notes/) to keep agents out of your personal notes. Updating frontmatter properties is always allowed regardless of folder restrictions.
• No destructive operations — agents cannot delete, rename, or move notes. The write tools only add content; they never replace or remove existing text.

Provenance tracking

Every write an agent makes is recorded in an audit log: which tool was called, which session called it, which note was affected, what was written, and whether it succeeded. These records are searchable through the unified Vault search. Read access is also tracked (aggregated by note and session, with 90-day retention).

Unified Search (Ctrl+P inside The Vault)

The right rail's Search panel opens a Command-Palette-style modal that searches three things at once:

• Notes — Files: top 8 matches against note titles + file paths.
• Notes — Content: top 12 matches against note bodies (FTS5 bm25-ranked) with snippet previews wrapping the matched terms.
• Images: top 8 matches against AI-derived image metadata (title + description + OCR text + tags + category). Click an image hit to open it in the Lightbox.

You can narrow with filter chips in the query bar:

• tag:foo — restrict to notes tagged #foo.
• file:segment — restrict by path substring.
• hidden:true / hidden:false — include / exclude hidden notes (default excludes).
• since:2026-05-01 — only notes edited on or after the date.
• kind:note — show only Notes (skip Images).
• kind:image — show only Images (skip Notes).

Chips combine: kind:image tag:diagram pcb finds analyzed images whose AI metadata mentions "pcb" and whose note context is tagged #diagram.

Watching the vault from other apps

Omniscio's file watcher picks up external changes live. If you edit a note in Obsidian, VS Code, or Notepad while Omniscio is open, the file tree refreshes and (if the file is the active tab) the editor reloads — but only if you don't have unsaved local changes. Omniscio will never clobber an in-progress edit.

Conversely, Omniscio writes plain `.md` files that any other Markdown app can open. The Vault does not "own" the format — it round-trips your text through a standard Markdown serializer.

The Vault in its own window

You can open The Vault in a separate floating window — useful for keeping your notes visible on a second monitor while Omniscio's main window stays on its primary view. Four ways to open it:

• Pop-out button — click the ExternalLink icon in the Vault toolbar's View group (tooltip: "Open The Vault in a separate window").
• Sidebar right-click — right-click The Vault row in the hubs sidebar and pick Open in new window.
• Global hotkey — press Ctrl+Alt+N from anywhere on your desktop, even when Omniscio is in the background. Customizable and toggleable under Settings → Keyboard Shortcuts ("Vault window" row).
• Tray menu — right-click the Omniscio tray icon and pick Open The Vault.

The standalone window shows the same file tree and editor as the Vault panel inside Omniscio — both surfaces read and write the same vault files and stay in sync live. Opening the window a second time brings the existing window to the front rather than opening a duplicate. Omniscio remembers the window's size and position between opens, and the window follows your light/dark theme automatically. To save vertical space, your open note tabs sit right in the window's title bar — there is no separate “The Vault” bar above them. Press Ctrl+W / Cmd+W to close the window; your last typing is saved first.

Session Seeds (briefing + project context)

Two ways to give a Claude session a head start with your notes:

• Multi-note briefing — select 2–5 notes in the file tree (Ctrl+Click or Shift+Click), right-click, and pick Start briefing. A small dialog lets you optionally type a question, then Start Briefing launches a vault-scoped session with every selected note’s full body loaded as context. The AI can read and reference all of them from the first message.
• Project context notes — right-click a single note and pick Pin to <project> (one row per real project you have in Omniscio). That note becomes standing context for the project: every new session spawned in that project automatically receives the pinned note’s current body, no manual selection needed. You can pin up to 5 notes per project.

Living context — because note bodies are read fresh at session-launch time, an updated note automatically appears in the next session. No re-pinning or re-selecting needed.

Privacy & safety

• Omniscio only writes inside your vault root. Every save / rename / move / delete is path-validated server-side — absolute paths, .. traversal, system paths, and symlinks are rejected before any file touches disk.
• Soft-delete is non-destructive. Deleted notes go to .trash/ with a timestamp prefix; deleted images go to .trash/assets/. Recover by hand.
• Images load through a sandboxed protocol. The custom nothari-asset:// scheme is the only way vault images reach the editor — path traversal, symlinks, and non-image file types are all rejected at the protocol handler. SVGs get extra Content-Security-Policy hardening.
• External edits don't clobber your typing. The watcher only reloads the editor when your local copy is clean.
• No LLM calls unless you trigger one. The Vault surfaces that spend money — image analysis (Analyze / Analyze all), note summaries, and the Ask AI vault chat — each run only when you invoke them. Agent tools (read and write) hit local SQLite only — zero API cost.

Knowledge Aggregation

Knowledge Aggregation is an opt-in background worker that periodically reviews your completed AI sessions and writes summary notes into your Vault. It produces daily and/or weekly "Agent Insights" notes that capture decisions made, patterns discovered, bugs found, lessons learned, and open questions across all your sessions.

How it works

1. Collect — the worker queries sessions that ended during the period. This stage is free (database queries only).
2. Deep-dive — the most notable sessions (scored by turn count, engagement, cost, and keyword matches) are sent to a lightweight AI model for structured extraction. A configurable cost cap stops extraction early if spending exceeds the limit.
3. Synthesize — one AI call merges all extractions into a single Markdown note, written into your vault under an aggregation/ folder organized by period and project.

Turning it on

1. Make sure The Vault is enabled and a vault folder is configured (see above).
2. Open Settings → Features and flip Enable Knowledge Aggregation.
3. Choose whether to generate daily notes, weekly notes, or both, and set the preferred hour/day.
4. Optionally adjust the cost cap (the maximum spend per aggregation run) and the deep-dive limit (how many sessions get the full AI extraction).

Run Now

You can trigger a one-off aggregation run at any time via the CLI (POST /kms/aggregation/run) or the IPC channel. "Run Now" bypasses the cost cap so every notable session gets deep-dived regardless of spend.

Alerts

The worker raises an inbox alert if it hits the cost cap (so you know some sessions were skipped) or if a run fails entirely. Each alert links to Vault settings so you can adjust the cap or check your API key.

Update Vault (/update-vault)

A built-in skill that scans your sessions and logs what they did into your vault. Type /update-vault in any session to run it.

Two modes

• This session (default) — extracts findings from the current session and updates the matching vault note before you archive it.
• Batch scan (say “scan all” or “daily scan”) — scans every session from the last 24 hours across all projects in one pass. The lookback is configurable (24h, 72h, or 1 week).

What it captures

PRs shipped, bugs found and fixed, features built, test results, UX improvements, session costs, and blockers. Each session’s findings go into the matching feature note (QA Tracker) or a project note (Projects/ folder). The cross-feature Bug Log and the QA Dashboard are updated too.

During onboarding

The onboarding cascade introduces the skill after the daily-briefing mission: “Your agents’ work adds up fast. /update-vault is how you stay on top of it.”

Web Clipper

Clip any web page by URL — or a screenshot by OCR — into your vault as a clean Markdown note. Notes land in a dedicated Web Clips/ folder with YAML frontmatter (source URL, timestamp, tags).

From Quick Launch

Open Quick Launch (Ctrl+Space) and switch to the Web Clip tab (scissors icon). The URL field auto-detects if your clipboard contains an http:// or https:// URL and pre-fills it. Optionally add comma-separated tags. Press Enter or click Clip page. The modal closes instantly; clipping happens in the background. If it fails, a toast appears in the main window.

From a screenshot

Capture a screenshot with the snip tool (Ctrl+Alt+S). In the annotation editor, click Save to vault (only visible when KMS is enabled). The AI reads the screenshot, extracts all visible text, generates a title, and saves it as a vault note tagged screenshot and ocr. This uses one Anthropic API call.

From the CLI

Agents and scripts can clip pages programmatically: POST /kms/clip with a JSON body containing url (required), plus optional tags, title, and vaultId. Returns the created note.

Limits

• Only http:// and https:// URLs; private/local addresses are rejected.
• Only HTML pages (not PDFs, images, or other content types).
• 10 MB response limit, 15-second fetch timeout.
• If article extraction fails (e.g. JavaScript-heavy pages), falls back to plain text.

What's still missing

• Embedding / semantic search. Both note and image search are bag-of-words FTS5 today.
• Multiple vault roots. Omniscio supports exactly one vault at a time.
• Live updates on web / mobile. The Vault panel now has a dedicated phone layout (a slide-over Notes list, swipeable tabs, and a Format sheet with every control), but the web companion does not yet receive live cross-device updates — it shows your latest edits on the next load or note reopen.
• Migration UI for importing notes from the Vault desktop database.

Super Prompts are curated, high-quality prompt templates that you can use to kick off AI sessions. Instead of writing prompts from scratch, pick a Super Prompt and let the template do the work.

Using Super Prompts

1. Open the ⋮ menu at the top of the sessions sidebar (available in every project) and click Super Prompts — or click the Super Prompts icon in the toolbar, or press Ctrl+Shift+K anywhere.
2. Browse the available prompts by category, or search by title, tags, or description.
3. Select a prompt and click Launch (or press Enter) — Omniscio starts a new session pre-populated with the prompt body. By default the session launches into the project you are currently viewing; use the "Launch into" dropdown in the picker header to choose a different project, including the Claude virtual workspace.
4. The first agent turn begins immediately with your prompt already sent.

Remote Super Prompts

Omniscio can fetch Super Prompts from a remote server, so new prompts are available without updating the app. These are synced periodically in the background.

The Weekly Summary is the once-a-week sibling of the Daily Digest. Every Monday at 7 AM local (or whatever hour you pick), Omniscio recaps the just-completed ISO week (Monday through Sunday) and surfaces it as two cards in your Briefings inbox: a slim recap card (a scannable highlight of your week plus the by-the-numbers stats) and a separate “N suggestions from last week” card that lists that week's personalized coaching tips. Splitting them out means the suggestions get their own labelled row instead of sitting at the bottom of a long recap where they were easy to miss. As of May 2026 the Weekly Summary ships as a briefing inside the Daily Digest "Briefings" surface — same sidebar entry, same inbox section, same triage shortcuts. There is no separate Weekly Summary sidebar entry anymore; daily digests and weekly summaries appear side-by-side in the merged Briefings list, sorted by date.

Weekly Summary replaces the Workflow Coach. The same six privacy-preserving detectors that powered Workflow Coach are now inputs to the Weekly Summary prompt, so you still get the personalized "you sent five near-duplicate replies this week, consider saving a quick reply" guidance — you just get it once a week in one place instead of in a separate panel.

What you see

The recap card

• A scannable recap — a short bold headline plus 3-5 highlight bullets written by Claude Haiku, each calling out one real number or project (your most active project, sessions run, total spend, a notable week-over-week change). It leads with the highlights instead of dense paragraphs, so you can read your week at a glance.
• Stats card — sessions total / active, agent vs. operator message counts, top 5 projects by session count, top 8 features by usage count, weekly AI spend in USD, and total tokens for the week (combined input + output, with an in/out breakdown). Large figures are shown rounded — e.g. 25.4K messages, $237.6K spend, 62.2B tokens — so the card stays scannable; small counts (like active sessions) stay exact.

The suggestions card

• A counted row — its inbox row shows how many tips are waiting (e.g. “3 suggestions from last week”), so you can see there's something to act on without opening it.
• Suggestion cards — 3-5 personalized tips. Each has a title, a short why + how, and (when applicable) an action button: create-automation, archive-idle-sessions, or explore-feature. (Saving a phrase you repeat as a quick reply is now its own always-on inbox nudge, separate from the weekly digest.)
• Dismissing a suggestion — tap the X on a tip to open a five-chip why-picker: already doing it, not useful, not relevant, wrong, or other. Your reason is remembered and fed back into the next Monday's generation for 90 days, so the same kind of suggestion stops reappearing. Each dismissed tip collapses into a strip with an Undo link, and once you've handled them all the whole suggestions card clears itself from the inbox.

First-launch backfill

Weekly Summary is on by default, so the first time it runs Omniscio generates up to four weeks of backfill summaries in one shot (oldest first) so you don't open the feature to an empty list. Each week is fully gated by the same cost cap as the live path; if the cap is hit mid-backfill, the loop halts and resumes the next day. The backfill runs exactly once — even if it crashes partway, the "I already tried" flag is set so a transient error doesn't loop on every relaunch.

Inbox cards, snooze, archive

Each week lands as two rows in the unified Inbox under the Briefings section (the same section that hosts daily digests — both cadences share one inbox home as of May 2026): the recap card and the suggestions card. The recap also shows in the Briefings sidebar's Active bucket. The two are independent — dealing with one never removes the other. Archive or snooze the recap with the usual paths: middle-click the row (desktop), right-click → Archive, press E while the row is selected, or tap the X button shown only at mobile widths; Snooze and Unsnooze are in the right-click menu and the S hotkey opens the Snooze palette. The suggestions card is dismissed the same way (its Dismiss button, middle-click, or Snooze), and it also clears itself once every tip in it has been handled. Older recaps (>12 weeks) automatically fold into the Archived bucket.

Privacy

Weekly Summary never reads the raw text of your messages. The behavioral detectors emit anonymized hashes and counts only — the same canary test that pinned Workflow Coach's privacy guarantee continues to enforce this contract. The stats card is aggregate counts and totals from your database; the LLM prompt is built from those plus the detector hashes. Your conversation content stays private.

Cost control

Weekly Summary is on by default and uses a small amount of AI credits each Monday morning — but only once your active account is an API-key account. An account signed in with a Claude login (OAuth) can't call the underlying API, so the Monday run simply does nothing until an API key is available — you're never charged for a run that can't happen. There's a daily spending cap (default $0.10) that you can adjust in Settings → Email & Summaries → Daily digest → Weekly Summary, range $0.05 to $5.00. Once the cap is hit on any given day, generation pauses until the next day — including the manual Run now button.

Settings

Settings → Email & Summaries → Daily digest → scroll to the Weekly Summary sub-section at the bottom. Weekly Summary is on by default; use the master switch here to turn it off (or back on), pick a generation hour (0-23, default 7), and tune the daily cost cap. The Weekly Summary toggle is independent of the Daily Digest toggle — you can enable one without the other. Existing Workflow Coach users keep their toggle state and cap value — Omniscio migrates them automatically on first launch after the upgrade.

Daily Journal adds a gentle daily nudge to your inbox: run a short journaling check-in with your AI coach (a weekly review runs on the longer cadence too). It's a lightweight way to keep a writing habit without remembering to open a journal yourself.

How It Works

• Daily nudge — a short check-in row appears in your inbox, inviting you to journal for a minute or two.
• Weekly review — on the weekly cadence the coach prompts a slightly longer reflection over how your week went.
• Private to coaching — what you write stays inside the coaching feature and never leaks into your other sessions.

The Toolchain Manager helps you install and manage the development tools that Omniscio and your AI agents need to work effectively.

What It Does

• Detects installed tools — Checks which tools are available on your system
• Installs missing tools — Can automatically install tools you need
• Checks versions — Verifies tools are up to date
• Update alerts for Claude Code and Codex — Puts a card in your inbox with a one-click Update now when either falls behind; the update runs in a terminal on your computer, never in the background

Accessing the Toolchain

The Tools virtual hub in the sidebar shows all detected and available tools. You can also access it from Settings → Connected Tools.

Skills are specialized instructions and workflows that enhance what your AI agents can do. They're like training modules — install a skill, and your agents gain new abilities.

User Skills

User skills are stored in your ~/.claude/skills/ directory. You can:

• Browse installed skills
• Enable or disable individual skills
• Install new skills from the marketplace

Beautify Site

Beautify Site is a bundled skill that takes a web page or app screen that looks bland, generic, or obviously machine-made, and turns it into something deliberate. It is off by default — switch it on under Settings → Features → Enable Beautify Site skill and it installs at the next launch. After that there is no panel and no button — you just ask for what you want in plain language, and it starts.

Say something like "make this landing page look expensive", "this page looks AI-generated", or "redesign this UI", and the skill runs four steps in order: it picks one design direction and writes it down so the rest of the work follows it, chooses or builds components that fit that direction, adds motion that helps you understand the page rather than motion for its own sake, and finally audits the finished page and strips out the tell-tale signs of AI-generated design.

That last step is the point of the skill. Left alone, an AI reaches for the same handful of moves every time — a colour gradient poured into the headline letters, a small letterspaced label stapled above every heading, a fake terminal window in the hero typing itself out. Each is individually defensible and collectively a fingerprint. The skill treats those as banned and rebuilds the element instead of just recolouring it.

It designs boldly by default, and varies the look every time, because settling into one recognisable style is its own giveaway. It only holds back if you ask for something quiet or the project's own design notes say so — and when those two conflict, it tells you and lets you pick rather than quietly playing safe. What it never trades away for boldness: reduced-motion support, keyboard focus outlines, tap-target size, colour contrast, and no sideways scrolling on a phone.

Dev Pipeline

Dev Pipeline is an opt-in bundled skill that runs a full software-development workflow end to end on a single task — and stops to ask you at five checkpoints along the way, so a long autonomous run never gets ahead of your judgment. You describe the task once; it sets up an isolated workspace, investigates, plans, red-teams its own plan, builds with tests, tidies the code, updates the docs, and gets a branch ready to merge — pausing at each gate for your call.

Turning it on

It ships with Omniscio but is off by default. Enable it under Settings → Features → "Enable Dev Pipeline skill". Flipping it on installs the skill into ~/.claude/skills/dev-pipeline/; flipping it off removes it again. The change takes effect immediately — no restart needed.

Optional companions

Once the skill is on, five optional switches appear right under that same toggle. Each is independent — turn on only what you want:

• Gate auto-approval, per phase — a switch for each of the five gates (Plan, Red Team, Build, Elegance, Docs) for whether Omniscio approves it for you or always asks, plus a select-all toggle to flip them together. By default the Plan gate asks you and the other four approve themselves when a phase reports done; a gate that asks a question or hits a blocker is never auto-approved.
• Remind every session to use the pipeline — adds a short note to every session telling it to follow the pipeline when it's doing software development, and to ignore it otherwise. Skips the in-app helper bots. Off by default.
• Add the pipeline quick replies — adds the "Full Local Git", "Open a Pull Request", "Package My Work Into PRs", and "Worktree Cleanup Analysis" quick replies if they're missing, tucked into a "Dev Pipeline" folder at the top of your list. They follow the order you actually work in: "Full Local Git" (a one-tap local land-all-ready plus worktree/branch cleanup), "Open a Pull Request" (finish one worktree and open its pull request), "Package My Work Into PRs" (take everything your checkout has that the shared copy doesn't and split it into the fewest clean, easy-to-review pull requests — it shows you the plan and waits for your go-ahead before opening any), and "Worktree Cleanup Analysis" (tidy up the leftover working copies the first three leave behind). Turning it off never deletes ones you already have, and a reply you already keep is left where it is. Off by default.
• Add the workflow auto-replies — adds a small set of preset auto-reply rules for the pipeline's common handoffs (for example, answering a finished red team with "proceed"). Two-way: turning it off removes exactly the presets it added, never rules you wrote yourself. Off by default.
• Write a durable note into the repo — keeps a short, clearly-marked block in the repository's own CLAUDE.md (or AGENTS.md, or it creates CLAUDE.md) telling sessions to route development work through the pipeline. Unlike the reminder above, this one is written into a file you can see and commit, so it also reaches sessions Omniscio didn't start. Two-way: turning it off strips the block out again. Off by default.

Running it

In any Claude Code session, start a run with the task you want done:

/dev-pipeline "the task you want done"

(or just say "run the dev pipeline on X"). If a run is interrupted — you close the session, or it compacts — invoke it again and it resumes from where it left off rather than starting over.

The six phases and five gates

The work runs in six phases. Phases 1–5 each end at a hard stop — a gate where you reply approve, give feedback to revise, or abort. Phase 6 finishes on its own.

What you do at each gate

• Approve — the pipeline advances and works autonomously until the next gate.
• Feedback — it stays on the current phase, applies your changes, and re-presents at the same gate.
• Abort — it stops and leaves the worktree and its notes intact for you to inspect.

Running it hands-free (autonomous mode)

If you'd rather not sit at every checkpoint, ask for it to run autonomously — say "run autonomously", "no check-ins", or "leave it overnight" (or start with /dev-pipeline --auto "the task"). It then drives itself through every phase, doing the full work of each, and only stops to ask when something genuinely needs you: a real decision (scope, a trade-off, anything hard to undo), a task too ambiguous to plan, a risky rewrite, or a build it can't get green. Everything else it approves itself and keeps going. The default is still the five-checkpoint flow above; autonomous mode is per-run, and because it's recorded in the run's notes, a resumed run keeps going hands-free.

QA Testing

QA Testing is an always-on bundled skill that automates QA handoff — composing a structured email from a customizable template and sending it to the VA team through Omniscio's approval-gated Gmail API. It's installed automatically into ~/.claude/skills/qa-testing/ on every app launch, so any session can use it.

Using it

In any Claude Code session, invoke it with:

/qa-testing

(or say “send this to QA”, “hand off to the VA team”, or “request QA testing”). The skill walks you through an 8-step interview:

1. Auto-detect PR info from the current branch
2. Confirm the feature name, PR number, and URL
3. Ask what to test — areas and flows the QA team should focus on
4. Ask for setup instructions — anything the team needs before testing
5. Compose the email from your customizable template
6. Show the draft and wait for your approval
7. Send via the CLI endpoint — queued for inbox approval
8. Report the result

Customizing the email

Edit ~/.claude/skills/qa-testing/email-template.md to change the default recipient, subject format, SOP link, or email body. The skill reads it fresh each time. Omniscio updates the skill instructions (SKILL.md) on launch but never overwrites a template you've customized.

Approval gating

The send is approval-gated through the same system as Gmail sends — it lands in your Omniscio inbox for one final confirm before the email goes out. Controlled by Settings → CLI Control → Require approval for Gmail sends (on by default).

Mind Map Generator

Mind Map Generator is a bundled skill that creates well-structured mind maps from your source material — vault notes, local files, or Google Docs. Instead of dumping raw data into the Mind Map feature (which can produce a flat, unusable list), the skill interviews you about scope, angle, and depth, then synthesizes a properly structured map within node and depth budgets.

Using it

In any Claude Code session, invoke it with:

/mind-map

(or say “make a mind map from my vault notes about X”, “mind map from this file”, or “visualize as mind map”). The skill walks through these steps:

1. Feature gate — checks whether Mind Map is enabled; if not, turns it on automatically (or directs you to approve it in your inbox)
2. Source resolution — reads vault notes, local files, or Google Docs
3. Interview — asks up to 4 questions (source, angle, depth, preview), skipping any you already answered
4. Synthesis — builds a structured outline with proper branching, short labels, and rolled-up counts
5. Validation — six checks: node count, depth, label length, single-child branches, fat branches, root format
6. Preview — shows you the outline if you opted in; you can approve, request changes, or start over
7. Creation — imports the map and gives you a link to open it

Depth tiers

When the skill asks how detailed your map should be, choose one of three tiers:

• Overview — major themes only (15–25 nodes, 3 levels). Good for presentations.
• Standard — key detail with some rollup (25–40 nodes, 4 levels). The default.
• Detailed — full depth when you need it (40–100 nodes, up to 10 levels). You must request this one.

Hard ceilings: 100 nodes and 10 levels. These are never exceeded unless you explicitly ask for more.

Skills Marketplace

Discover and install community-created skills. Skills cover areas like:

• Code review workflows
• Testing strategies
• Documentation generation
• Debugging approaches
• And many more

Stored Keys is one place to see every API key and secret Omniscio is aware of, grouped by where it's kept. It shows names and metadata only — never the secret values. Nothing is decrypted to build the list, and nothing is decrypted when you manage a key. From here you can search, remove, and replace keys without ever exposing them.

Where it is

Open Agent Tools → Stored Keys in the sidebar — alongside CLI Tools, Skills, and MCP Servers.

What it shows

A search box at the top filters live by name and type. Entries are grouped into four sources:

Each row shows the name, a type label, and a last-updated date where one is recorded. A Refresh button re-reads the inventory. Empty sources are omitted; an empty inventory reads "No stored keys found".

Managing keys

Every row can be removed behind a confirmation. What "change the value" does depends on the source:

• Windows vault — remove inline (the encrypted file moves to a hidden trash so you can Undo for a few seconds); add or replace a value inline (re-encrypted on your machine). The Add secret button (top-right) creates a new vault secret — names are lowercase letters, numbers, and hyphens.
• Provider keys / Claude accounts / Automation credentials — remove inline; Manage → jumps to the relevant Settings editor to change the value. (You can't remove your only Claude account.)

Deleting a secret Omniscio itself relies on (such as its own command-server key) makes you type its name to confirm, so you can't break Omniscio by accident.

When you paste a secret into the session composer and the feature is on, Omniscio saves it in an encrypted per-session store on your machine. On Claude-based engines it puts a short placeholder — like AMCSECRET-a3f9b2… — in its place; that placeholder is all the AI ever sees, and when the agent writes a shell command, creates a file, or fills in a web form containing it, Omniscio swaps the real value back in privately, at the moment of execution, so the secret reaches its destination without entering the conversation. On other engines the value still reaches the engine (so its tools keep working), but Omniscio keeps it out of your saved history and tells you so plainly.

How to use it

1. Enable the feature at Settings → Features → "Secret Paste Guard" (secretHandleEnabled).
2. Paste your API key or secret into the session composer as you normally would. Omniscio detects that it looks like a secret and replaces it with a handle automatically.
3. Give the agent its task. The agent reads the handle, uses it in whatever commands or files it needs, and Omniscio substitutes the real value at execution time.
4. Disable it any time by toggling secretHandleEnabled off. After that, pasted text is treated as plain text again.

What it protects

Omniscio’s own saved history is scrubbed of the value on every engine — the raw secret is never written to:

• Your Omniscio conversation history or database
• The full-text search index
• Session exports
• The Claude CLI’s own local transcript (on Claude-based engines — a log-scrub cleans the value back to its placeholder once the session ends)
• The live chat window (even if the agent tries to echo it back — a scrub step removes it before the output is saved or displayed)

On Claude-based engines the model never sees the value at all. On other engines the value still reaches the engine and its own logs — only Omniscio’s saved history is protected (see Scope below).

Scope

• Fully hidden (Claude-based engines): the native Claude engine and the vendors that reuse it — the model never sees the value, and the local transcript is cleaned. SSH/remote is out of scope.
• codex / gemini / cursor: server-side secret tools are a planned fast-follow; until they land, these run with partial protection.
• Every other engine (partial): Omniscio’s saved history is scrubbed, but the engine still receives the key — the offer says so honestly.
• Web-form fill via the AI Browser resolves the value privately, server-side.
• Desktop only. The feature is not reachable from the mobile or web interface — blocked by design so a headless endpoint can’t be used to inject secrets remotely.

Where secrets are stored

Each secret is encrypted using your operating system’s own keyring (Windows DPAPI / macOS Keychain / Linux libsecret) and saved in a per-session file on your machine at <userData>/session-secrets/. If the OS keyring is unavailable, the value is kept in memory only — nothing is written to disk as plaintext. The store is wiped when the session is deleted or reaches its retention limit.

Every AI coding agent reads its instructions from a slightly different file. Claude Code reads CLAUDE.md. Codex, Cursor (modern), Gemini, and Factory have settled on AGENTS.md. Gemini CLI reads GEMINI.md. Cursor's legacy file is .cursorrules. Cline reads .clinerules/base.md. GitHub Copilot reads .github/copilot-instructions.md. Windsurf reads .windsurfrules.

If you work in a project with more than one agent — even just Claude Code plus your editor's Copilot — you either have multiple copies of the same rules to maintain by hand, or one of them is stale.

Agent Instructions Sync collapses that into a single source of truth. You pick one file as your canonical (default: CLAUDE.md), check off which other filenames you want to keep in sync, and Omniscio automatically copies the canonical's contents — and, for a project that has one, the rule files under .claude/rules/ — to every mirror, whenever you edit the canonical, when an agent edits it during a session, or before any new Claude CLI session starts.

The Seven Recognized Files

Turning It On

1. Open Settings → Features → Agent Instructions Sync.
2. Flip the Enabled toggle on.
3. Pick your canonical file: CLAUDE.md (default) or AGENTS.md. Whichever you pick becomes the source of truth; the other becomes a mirror you can opt into.
4. In the Mirrors list, check off the other filenames you want kept in sync.
5. Click Sync now to run the first sync immediately. The selected mirror files appear in your project's root directory, each starting with a small Omniscio banner comment.

Projects that don't have a canonical file present are silently skipped — nothing is created from nothing.

The Banner: Why It's Safe

Every mirror file Omniscio writes gets a small HTML-style comment at the top:

<!-- AMC-SYNC: synced from CLAUDE.md at 2026-05-14T03:35:08Z — do not edit, edits will be overwritten -->

Omniscio will only ever re-touch a file that has this banner. A .cursorrules you wrote by hand before turning sync on is never silently overwritten — Omniscio sees no banner, marks the row "manual file" in the sidebar, and leaves it alone. Same for any future external edit: the body hash changes, Omniscio notices, and asks you what to do instead of writing over it.

What Triggers a Sync

Three ways:

• Manual — click Sync now in Settings, or click the manual sync button in the sidebar.
• File watcher — Omniscio watches your canonical file, and the project's .claude/rules/ directory if it has one (those rule files are loaded on every turn just like CLAUDE.md, and the other agents never read them, so a mirror has to carry them). Edit either one (in Omniscio, in VS Code, an agent edits it during a session) and a sync runs about half a second later. An echo guard makes sure Omniscio's own writes don't loop back as new edits.
• Pre-spawn hook — every time you launch a Claude CLI session, Omniscio re-syncs first. Fire-and-forget — never blocks the spawn.

The Sidebar Surface

Open any project. In the right-hand Dashboard sidebar, expand the Auto Context (N) panel and find the System Instructions sub-group (the rows that list project CLAUDE.md, project MEMORY.md, etc.). Below those rows, if you've enabled sync and the project has any synced mirrors, you'll see a small Synced mirrors subheader with one row per mirror:

• Green dot — "in sync": file matches the canonical. Click the row to peek at the contents.
• Blue dot — "out of sync": the canonical has moved on, but the mirror still holds exactly what Omniscio last wrote into it. Auto-refreshes on the next sync pass — nothing for you to do. Informational only; not clickable.
• Amber dot — "edited externally": Omniscio has a recorded baseline for the mirror and something has changed it since Omniscio last wrote it (a real edit, not just a stale copy). Click the chip to open the conflict resolver (see below).
• Muted gray dot — "still tracked in git": in the default "keep copies out of git" mode, the copy looked edited but is really just still tracked in git — a branch merge keeps restoring the old committed version over Omniscio's fresh one. You didn't edit it; git did. Click the chip to fix it in one step (see below). This replaces the alarming "edited externally" flag for this specific, common case.
• Muted gray dot — "needs first sync": the mirror's body differs from the canonical, but Omniscio has no recorded baseline for it yet — the day-zero state on a fresh install, after the sidecar state file was cleared, or for mirrors written before Omniscio started tracking baselines. Omniscio can't tell "I wrote this long ago" from "you edited it," so it refuses to overwrite and asks you to pick once. You haven't done anything wrong; it's a one-time setup step. Click the chip to open the same resolver — only the wording is neutral instead of accusatory.
• Gray dot — "manual file": a file with the right name exists, but it has no Omniscio banner. Omniscio won't touch it.

Resolving an "Edited Externally", "Needs First Sync", or "Still Tracked in Git" Conflict

When Omniscio won't auto-overwrite a mirror — either because some other tool changed it since the last sync ("edited externally"), or because Omniscio has no recorded baseline for it yet ("needs first sync") — clicking the chip opens a dialog with the same three options below. Only the title and explanatory copy differ between those two cases; the accusatory wording is dropped for "needs first sync" because you haven't done anything wrong. Default focus is Cancel, so pressing Enter never overwrites a file by accident.

The third case, "still tracked in git", is different: the copy isn't really edited — git keeps reverting Omniscio's writes because the file is still tracked. So that dialog shows a single button, "Stop tracking in git", which runs git rm --cached (the file stays on your disk) and ends the problem for good. Discarding or promoting wouldn't help — the next merge would just revert it again.

1. Discard edits (red) — force-overwrites the mirror with the current canonical's body. The external edits are gone. Use this when the external edit was a mistake.
2. Promote to canonical (amber) — reads the mirror's body, strips the banner, atomic-writes the result over the canonical file. The external edit becomes the new source of truth for everything going forward. Use this when the external edit was actually the improvement you want everywhere.
3. Stop syncing this mirror (ghost) — removes the mirror from your sync list. The file on disk stays exactly where it is; Omniscio just won't write it again.

Turning It Off Without Losing Files

Flip the Enabled toggle off. The watcher stops; no new writes happen; no existing files are deleted. Every mirror file already on disk stays exactly where it is, banner and all. If you turn the toggle back on later, Omniscio re-probes — files that haven't changed go straight back to "in sync"; files that have changed during the off-period show up as "edited externally" and wait for your decision.

How Copies Are Handled in Git

The mirror files are generated from your canonical file, so committing them to git is awkward — a file Omniscio keeps rewriting churns on every merge and can leave a stale committed copy that trips a false "edited externally" flag. Omniscio gives you an explicit choice under Settings → Features → Agent Instructions Sync → "How copies are handled in git":

• Keep copies out of git (recommended, the default). Omniscio adds the mirror filenames to your project's .gitignore (in a clearly-marked block) so git never tracks them. They're rebuilt on each machine from the canonical, so a merge can never clobber them. If you already committed some copies, the panel shows a one-click "Stop tracking copies in git" button that runs git rm --cached on each — it removes them from git's index but keeps the files on your disk. If you never open that Settings panel, the same one-click fix now also appears right on the sidebar: a still-tracked copy shows a calm "still tracked in git" chip that opens the fix directly.
• Keep copies in git. For repositories you share with teammates who use Codex/Gemini without Omniscio — they need the files physically present. Omniscio installs a small post-merge / post-rewrite git hook that rebuilds the copies from the canonical after a merge or rebase, so they never go stale. Omniscio won't touch a repo that manages its own hooks (e.g. Husky); there the copies stay merge-safe thanks to the stable banner.

Long agent replies are exhausting. Plain Speak takes the agent's wall of text and rewrites it as a small, predictable structure: a one-line summary at the top, a clear recommended action, the original response below, the latest progress, and any open questions. You see the same five sections on every reply, so your eyes land on the part you care about without re-reading.

The Five Sections

Every Plain Speak message gets rewritten into the same five sections, in the same order:

• TLDR — one line summarizing the whole message.
• Recommended Action — what Omniscio thinks you should do next.
• Response — the agent's actual reply, lightly tidied.
• Latest — the most recent progress or status line.
• Questions — any open questions the agent asked you, rendered as Question Widget pills you can answer with one click.

If a section has nothing to put in it — no recommended action, no open questions — that section is hidden entirely, so you never see an empty "(none)" placeholder row. And if a whole message has nothing worth summarizing, Plain Speak hides the card and shows the agent's original reply instead.

Turning It On

1. Open Settings → Plain Speak.
2. Flip the master Enabled toggle on.

With the master toggle on, Omniscio processes every new agent message through Plain Speak as it arrives — on every session, no per-session opt-in. (The per-session opt-out checkbox in the session three-dot menu was removed on 2026-05-14 to reduce menu clutter; as of 2026-05-26 the master toggle is the single on/off control — the separate pause flag was retired so the toggle can never say "on" while nothing fires. The only control now is the global master enable — the agent-written card is free, so the old daily cost cap was removed from the panel.) Older messages in the transcript are left alone unless you explicitly re-trigger them.

The V Hotkey

Press V while focused on a session to toggle between the Plain Speak version and the original message body. The toggle is per-message and remembers your choice for that message until the session is closed. If you want to read what the model actually said, V is the fastest way.

Overlay Pending

Plain Speak runs as a background task after the agent's message has already appeared. While it's running, Omniscio shows the original message immediately and stamps a small Plain Speak pending overlay. As soon as the rewrite completes, a Plain Speak pill appears in the message header; by default the body stays on the original reply with the card one click away (unless you turned on Show Plain Speak first). Either way you never wait on Plain Speak to read your agent's reply.

The session row in the sidebar shows a green dot during the overlay-pending state so you know an agent message arrived and is being rewritten. Once Plain Speak finishes, the dot clears and the session returns to whatever attention state it was in before.

You won't get yanked to the card mid-read. Because messages open on the original reply by default, the rewrite arriving never interrupts your reading — the body stays put and the Plain Speak pill simply appears. The same protection covers the Show Plain Speak first case: if you've turned that on, Omniscio checks whether you're actively looking at the bubble when the rewrite lands (the session is active, the window is in the foreground, and the bubble is in view), and if you are, it leaves the message on the original markdown you were reading instead of flipping to the card mid-paragraph — common when you have a green (running) session open and the rewrite drifts in 30–60 seconds later. You can press V or click the pill any time to flip it yourself.

When Plain Speak Doesn't Run

A few cases short-circuit the rewrite and leave the original message untouched:

• The master toggle is off.
• The message is very short — no point summarizing two sentences.
• The agent didn't include a card — an older session from before the feature, or a reply with nothing worth summarizing.
• The session is paused, archived, or otherwise terminal.
• The reply is a dev-pipeline status card — one of the 🔍/⚔️/🔨/💎/📝/🚀 gate cards (Plan Ready, Red Team, Build, Elegance, Docs, Ready to Merge). These are already laid out for you with their emojis and structure, so Plain Speak leaves them in that native format instead of flattening them into the five-section summary. They still alert you right away.

In any of these cases Omniscio renders the original message exactly as the model wrote it, and the V hotkey is a no-op (nothing to toggle to).

When an agent finishes a turn and flags itself as needing your input, sometimes that flag is right ("I'm blocked, please confirm the database name") and sometimes it's wrong ("here's the answer to your question, hope this helps" — not actually a question). Inbox Pilot reads each newly-arrived agent message with a fast small model and decides what to do with the session.

The Five Outcomes

For each newly-arrived agent message, Inbox Pilot picks one of five outcomes:

• Keep Needs You — the session really does need your input. Leaves the session in Needs You so you'll see it.
• Hide — the session is informational. Returns it to the regular section so it doesn't fight for your attention.
• Archive — the session is done. Archives it outright.
• Snooze — the session will be relevant later. Snoozes it for a model-chosen interval, then resurfaces it.
• Respond — the session needs a quick acknowledgement Inbox Pilot can send on your behalf (e.g. "yes, continue"). Posts the response and the session flows from there.

Turning It On

1. Click the Inbox Pilot tile in the left sidebar. If you don't see it, it is off on a default install — turn it on from the Projects sidebar's ⋮ menu → Sidebar Items, then flip Inbox Pilot on. The tile appears as soon as you do.
2. Flip the master Enabled toggle on.
3. Pick a model. Haiku is the default — fast and cheap.
4. Optional: set per-session rules. Pin a single outcome for a chatty session ("always Keep Needs You", "always Archive") if you want to override the model's classification.

Per-Session Rule

Sometimes you know a particular session should always take a specific outcome — a long-running scraper that should always snooze for an hour, a noisy notifier that should always be hidden. Right-click any session and pick an Inbox Pilot rule. From then on, every classification for that session uses your pinned outcome instead of running the model.

Pinned rules are per-session, persist across restarts, and survive session-end / re-open cycles. Clear a rule by selecting Use AI classification from the same context menu.

Test on Latest Message (Dry-Run)

Want to see what Inbox Pilot would do on a particular session before relying on it? Open the session's three-dot menu and pick Test on latest message…. Omniscio re-classifies the session's most recent agent message using both your global Inbox Pilot rule and any per-session pin, and shows you the outcome (Keep Needs You / Hide / Archive / Snooze / Respond) along with the model's reason.

• No action fires — the session is not archived, hidden, snoozed, or responded to. It's purely a "what would you do?" check.
• The run is logged in the Inbox Pilot activity history flagged as a dry-run, and is not counted toward your daily cost cap.
• Greyed-out with a tooltip when Inbox Pilot is off for this session — turn it on (master toggle or per-session toggle) to enable.

Use this when you're tuning your global rule or a session pin and want to confirm "are we on the same page?" before letting Pilot act automatically.

Snippet Replies (Pre-Approved Auto-Sends)

When Inbox Pilot picks Respond, it normally writes a short freeform reply ("yes, continue"). You can do better: hand it a list of your own Quick Reply snippets and let it pick one that matches the conversation. That keeps auto-replies on-brand, predictable, and cheaper — the classifier emits a short code instead of generating prose.

Which Snippets Are Eligible

For a snippet to show up in Inbox Pilot's options, it must:

• Be a regular snippet (not a folder, not a divider).
• Have Auto-submit on click turned on — edit-then-send snippets are excluded because they expect a human to tweak the text before sending.
• Contain no {variable} placeholders — Inbox Pilot has no way to fill them in.
• Have the per-snippet Allow Inbox Pilot to auto-send this reply toggle on in the editor's Behavior tab (it's on by default for new snippets).

Per-Snippet Opt-Out

Each snippet has its own Allow Inbox Pilot to auto-send this reply toggle on the editor's Behavior tab. Turn it off for snippets you only want fired by a hotkey or click — for example, a "Send my home address" snippet you don't want an AI choosing on your behalf. The toggle is a per-snippet kill-switch; the rest of the snippet keeps working everywhere else.

Helping the Classifier Pick Well

When the per-snippet toggle is on, an optional Hint for the classifier field appears underneath — one short sentence telling the classifier when to use this snippet ("confirm a yes/no question", "acknowledge a finished task"). When set, the classifier sees the hint instead of the snippet's body, which keeps the prompt compact and makes its choice much more accurate. If you leave the hint blank, Omniscio falls back to the first ~80 characters of the snippet text.

Audit Trail

Every snippet-fired Respond logs both the snippet's id and a frozen copy of its text at dispatch time. If you later edit or delete the saved snippet, the audit log still shows exactly what was sent. The snippet's usage counter is intentionally not bumped by auto-fires — that count is reserved for your own clicks and hotkeys, so it stays a meaningful "snippets I personally reach for" metric.

Cost

Inbox Pilot runs Haiku (or your chosen model) once per inbound agent message, with the message body as the prompt. That's a tiny request per turn, but it does add up across a busy day of many sessions. The Stats virtual hub's Feature Usage tab shows your total Inbox Pilot spend, and you can set a daily cap in the Inbox Pilot tile if you want a hard ceiling.

Daily Digest is an AI-generated briefing Omniscio produces once per day. It pulls together what changed overnight — new important emails, today's calendar events, sessions that finished or got stuck, anything an automation flagged — and writes a short summary you can read in under a minute over coffee.

The Briefings Sidebar

Daily Digests live in their own Briefings virtual hub in the sidebar, similar to Bookmarks or Automations. The Briefings panel has three tabs:

• Active — today's briefing and any from the last few days you haven't dismissed.
• Snoozed — briefings you snoozed for later. They resurface at the chosen time.
• Archived — everything you've dismissed. Kept for reference; the archive doesn't expire.

Hotkeys

• X — archive the focused briefing.
• Shift + X — snooze the focused briefing. Pick an interval from the snooze menu.
• Shift + D — jump straight to the Daily Digest panel from anywhere in the app.

In your Inbox

Unread briefings also show up in your main Inbox — one row per briefing, newest first, each labeled with its date. Open or archive a briefing right there like anything else in the inbox (archiving one clears just that briefing), or jump to the full Briefings panel above for the complete history.

What Goes Into a Briefing

Daily Digest pulls from every integration you've connected. If you haven't connected something, that section is just omitted — the briefing has whatever's available, never empty filler:

• Gmail — important new emails from the last 24 hours.
• Google Calendar — today's events.
• Sessions — anything that needs you, finished overnight, or errored.
• Automations & Recipes — anything an automation or cron-scheduled recipe flagged for review.

Turning It On

1. Open Settings → Email & Summaries → Digest & Summaries.
2. Flip the Enabled toggle on.
3. Pick a delivery time. Default is 7:00 AM in your local time zone.
4. Optional: choose which integrations to include. By default everything you've connected is in.

Drip is a queue-and-trickle inbox feeder. You stash short notes, links, files, or watched folders into a named drip, set a cadence, and Drip releases items into your inbox a few at a time. The metaphor is a slow drip rather than a firehose — perfect for “weekend reading,” “papers to read,” or a folder of resources you want walked through over a few weeks.

What a Drip Looks Like

Each drip is one named queue plus one cadence (e.g. “Weekend reading — Saturdays at 9am”), plus the pile of pending items waiting to be released. A drip has three statuses:

• Active — the schedule fires on cadence, releasing the configured number of items per tick from the head of the queue.
• Paused — you stopped it. Pending items are preserved, so resuming picks up where you left off.
• Archived — the queue drained naturally. A one-shot “queue empty” card lands in your inbox so the drip doesn’t just go silent on you. (A Repeat forever drip never reaches this state — see Per-Drip Knobs below.)

What You Can Queue

• Text — freeform prose or Markdown. Renders as rich text in the inbox viewer.
• Links — paste a URL and Omniscio fetches an Open Graph preview (title, description, hero image) so the inbox card looks like a clickable bookmark, not a bare URL.
• Files — attach any file. Omniscio keeps its own copy under the app’s data folder, so the original can move or be deleted later without breaking the drip.
• Folder sources — point a drip at a directory on disk. Omniscio walks the folder on a slow cadence and queues new files automatically. Useful for “trickle my Downloads/research folder back to me over the next few weeks.”
• Books — attach a PDF or EPUB and Drip releases it a unit at a time instead of as one file: a single PDF page, one EPUB chapter, or a fixed-size run of words for an EPUB whose chapters are too coarse. Good for trickling a whole book into your inbox over weeks.

Setting the Cadence

The cadence picker works like the Snooze palette — one input box with a list of presets beneath it that filters live as you type. No separate “Raw cron” tab, no red error on your first keystroke.

• Leave the input empty and pick from the seeded presets: Every day at 9am, Every weekday at 9am, Every Monday at 9am, Every weekend at 10am, Every hour.
• Type e to highlight every preset starting with “Every…”; type every fri to narrow to Friday entries.
• Type a full plain-English cadence the parser understands — “every weekday at 9am”, “every Saturday at noon”, “Monday and Thursday at 7pm” — and a Custom row appears at the bottom of the list with the resolved cron beside it.
• Paste a 5-field cron expression like 0 9 * * 1-5 and Omniscio detects it automatically — no mode switch needed.
• Use ↑ / ↓ to move through the list and Enter to pick. Omniscio previews the next few fire times beneath the input so you can confirm before saving.

How Items Reach the Inbox

When a drip fires, it releases items from the head of the queue (1 per tick by default; adjustable per drip). Each released item appears in your unified inbox as a Drip integration row with the drip’s name as the source label. Clicking the row opens the item in the Drip inbox viewer on the right:

• Text items render as Markdown.
• Link items show a clickable preview card — hero image, title, description, domain — that opens in your default browser when clicked.
• File items show a name / type / size pill.
• Queue-drained items show a one-line “all items released” confirmation.

The standard inbox actions work: snooze (push the item back into the queue or just delay it), archive (press E, or middle-click the sidebar row on desktop), or undo the archive from the toast that appears.

You can also right-click the content in the viewer, just like anywhere else on your computer: right-click a picture for Copy Image / Save Image As…, a link for Open Link / Copy Link Address, or text for Copy (no need to highlight it first). (Desktop only.)

Each item’s viewer also carries a Start session button (a message-plus icon beside Archive). Click it to spin up a Claude Code session that opens already focused on that item: Omniscio pre-fills the opening prompt from the item’s content — led by this drip’s saved preset, if you set one — and picks a target repo, both of which you can edit before confirming. By default the session opens in your Claude repo; pick any other project from the dropdown — now showing each repo’s icon — to launch there instead. The prompt box follows your usual send-key setting (Enter, or Ctrl+Enter, to send). (The button is hidden on the “queue drained” card and disabled if you have no repo to launch into.)

Made a typo, or want to reword a reminder after it arrived? Text notes are editable in place. In the inbox viewer, click the pencil (beside Snooze / Archive, shown for text notes only) to turn the note into an editable box — Ctrl+Enter saves, Esc cancels. You can also edit from the Drip management view: each text row has an Edit button, so you can fix a note even before it’s released. The change shows up everywhere the item appears — the inbox, the management view, and your phone. (This edits the note’s words; to change the drip’s schedule, click the drip’s name at the top of the inbox viewer.)

Archived items — how to find them, how to restore

Archiving a drip item just removes it from the inbox — it’s not deleted. To see archived items, open Drip in the sidebar, pick the drip the item belonged to, and switch to the Archived tab. Each archived row has a Restore button that puts the item back into the inbox (it’ll be there next time you open the inbox).

The four tabs in a drip’s management view are: Pending (waiting to be released), Released (already trickled to the inbox), Archived (released and then dismissed from the inbox), and All (everything regardless of state).

Deleting a drip — and what the cleanup removes on its own

Deleting a drip is permanent. The confirm dialog says so, and there is no undo. Behind the scenes the deleted drip sits for 7 days before an hourly cleanup job removes it for good — every database row and its whole on-disk folder, including any file copied in from a watched folder and any imported PDF or EPUB. There is no trash folder for a drip, so copy out anything you want to keep before you delete.

There is a second cleanup you never ask for. Inside every drip you still have, that same hourly job removes any file in the drip’s workspace that no live item points at and that has not been touched in 24 hours. You did not delete that file — the cleanup did. Two things are safe from it: a file still referenced by an item you archived (archiving is not a delete, so the item — and its file — stays), and the original file in a folder you asked Drip to watch. Drip always copies a file into its own workspace rather than moving it, so neither deleting a drip nor this cleanup can reach the folder you pointed at.

Per-Drip Knobs

• Repeat forever — when on, the drip never stops. Instead of trickling its queue once and archiving when empty, it re-delivers its item(s) into your inbox as a fresh card on every scheduled tick, indefinitely. Perfect for a recurring reminder — e.g. a photo or a note you want to see every day at 9am. Leave it off for the classic trickle-then-stop behavior. (For a multi-item repeating drip the items rotate one per tick; to pin a specific item to a specific time, use one repeating drip per time slot.)
• Items per release — how many items the scheduler releases each tick. Default is 1.
• Skip next on snooze — when on, snoozing a released drip item also skips the next scheduled release. Useful for buying breathing room without pausing the whole drip.
• Default session prompt — an optional preset that leads the opening prompt when you Start session from one of this drip’s inbox items. Leave it blank for no preset.
• Default repo — which repo a started session opens in. Leave it at Claude (default) to decide at launch, or pick a project to make that the saved default.

Turning It On / Off

1. Open Settings → Features.
2. Toggle Enable Drip. When on, the “Drip” row appears in the Omniscio group in the sidebar.
3. Click Drip in the sidebar to open the management view. Use + New drip to create your first one.

Inbox Alerts let an agent raise a standing item in your unified inbox whenever it has something worth your attention. The alert is persistent — it survives an app restart and lives until you archive or snooze it — so a long-running job can flag “the build is green” or “I need a decision here” and trust that you’ll see it the next time you open Omniscio, even if the session that raised it has long since finished.

What an Alert Can Carry

• Text — a short note or Markdown body that becomes the inbox row. Every alert’s text reads exactly like a chat message — headings, lists, links and code blocks included — whether an agent or Omniscio itself raised it, and whether you open it fresh or look it up later.
• A link — a URL that opens in your default browser when you click through.
• A file — Omniscio keeps its own copy, so the original can move or be deleted later without breaking the alert. If the file is an HTML page, it opens as a fully-styled page right inside the alert (see below) instead of a download chip.
• A “Start session” action — an optional button on the row that spins up a Claude Code session already focused on the alert, so you can act on it in one click.
• A project group — an alert can be filed under a project so related alerts cluster together in the inbox.

Repeats Collapse

An alert can include a key. If an agent raises another alert with the same key, Omniscio doesn’t stack a second row — it folds them into one row with a ×N count showing how many times it has fired. A chatty watcher that pings “disk almost full” every few minutes stays a single, quietly-incrementing line instead of burying your inbox.

No Repeat Nagging

Once you dismiss a problem or warning alert, Omniscio keeps that same alert from popping back for 24 hours — so a card like “low disk space” can’t reappear minutes after you clear it. It’s a blanket rule that covers every kind of problem alert at once, so you never have to quiet them one by one. Real messages are never held back: a new Team Chat message always comes through, and a genuine emergency (like runaway AI spend) still breaks through right away. Prefer the old behavior? Switch it off under Settings → Notifications → Limit repeat problem alerts (on by default).

A few alerts wait much longer than a day, because they describe a standing situation you either fix once or decide to live with — nagging you daily about a choice you already made just teaches you to dismiss them on sight. “Backups are on the same disk as your data” comes back at most once every 3 months; “Mobile Access is on” and the monthly AI allowance notice wait 30 days. None of them goes silent for good — each returns on its own schedule for as long as the situation lasts — and fixing the underlying thing clears the card straight away rather than waiting out the timer.

Browse Recently-Dismissed Alerts

Open the Alerts row in the sidebar and your current alerts are listed down the left, with the one you click open on the right — so you can work through a pile without losing your place in the list. It works the same in a popped-out Alerts window (Open in new window): click an alert and it opens beside the list, Go to session takes the main window to that session, and the list keeps itself up to date as alerts arrive, are snoozed or are dismissed elsewhere. Any button in the alert that leads somewhere else — a Settings link, Open channel, Start session — opens it in the main window and brings that window forward. A few that only work there, like View your records, open the alert itself in the main window, so one more click there finishes the job. Dismissed an alert and wish you hadn’t? Beneath your current alerts is a “Recently dismissed” list — everything you dismissed in the last 7 days, greyed out. Expand it and click any one to re-read it on the right, or hit Restore to move it back into your current alerts. After a week a dismissed alert drops off the list on its own. (Nothing is deleted early — your dismissed alerts were already kept; this just gives them a place to appear.)

The Alerts screen also has a Sessions tab beside Alerts. Start a session while you’re looking at your alerts — the + at the top, or New session on the tab — and it runs as the Alerts screen’s own agent, listed right there so you can find it again later. The tab shows a count when agents are running, waiting for you, or have hit an error, so you can tell at a glance without leaving your alerts; click one to open its chat, and click Alerts to come back to the list. It’s handy for handing an alert to an agent to investigate: start the session from the Alerts screen and it works out of the Alerts screen’s own workspace.

How Agents Raise One

An Omniscio-spawned agent posts to the local CLI control server’s /alert endpoint. The request is bearer-authed with your CLI token (read from the environment as $AMC_CLI_TOKEN); the agent supplies the alert text plus any optional link, file, key, project, or start-session action. Omniscio applies it immediately — raising an alert is benign and never spends Claude tokens, unlike spawning a session.

Who Raised It

Open an alert and its detail view shows a “Generated by <session>” link near the top — click it to jump straight to the agent session that raised the alert. Agents that Omniscio launched identify themselves automatically, so their alerts are traceable; an alert raised by hand or by an external script simply shows no origin. It is the same provenance trail Omniscio keeps for approvals and spawned sessions.

Broadcast response forms

Some inbox cards can also ask you a question. A broadcast from the Omniscio team can carry a short Broadcast response form — one or more questions (free-text or multiple choice) that you answer right on the card and submit back, without leaving your inbox. Your answer is saved locally first and then sent, so a brief network hiccup never loses it, and re-answering simply updates your previous response.

Broadcast read receipts

When the Omniscio team sends you a targeted in-app message, your app quietly confirms back to the sender that it was received (it reached your app), presented (it landed in your inbox), and read (you opened the card) — so the sender can see their message actually arrived. It is best-effort and privacy-respecting: nothing is reported if you turn off “Receive developer messages & surveys” in Settings, and public fleet-wide announcements are never tracked per person.